Market Organization and Market Efficiency in by ukg95843


									                Market Organization and Market Efficiency
                         in Electricity Markets

                          Erin T. Mansur†           and     Matthew W. White‡

                                              March 2007


      Electricity markets in the United States exhibit two different forms of organization:
      decentralized bilateral trading and centralized auction markets. Using detailed data
      on prices, quantities, and production costs, we examine how market outcomes changed
      when a large region in the Eastern US rapidly switched from a bilateral system of
      trade to a well-designed centralized auction market in 2004. Although economic theory
      yields ambiguous predictions, the empirical evidence indicates that shifting the venue
      of trade substantially improved overall market efficiency, and that these efficiency gains
      far exceeded implementation costs. Our analysis points to the merits of organized
      market institutions for electricity, a central issue in policy debates over market-oriented
      regulatory reforms.

     Yale School of Management, 135 Prospect St., P.O. Box 208200, New Haven, CT 06520-8200, Yale
School of Forestry and Environmental Studies, and NBER. Email:
     The Wharton School, University of Pennsylvania, Philadelphia PA 19104-6372, and NBER. Email:
     Acknowledgments. The authors gratefully acknowledge the University of California Energy Institute for
facilitating access to proprietary industry transaction data for this project. Special thanks to Joe Bowring
and the Market Monitoring Unit of the PJM Interconnection, LLC., for helpful discussions and information
about market procedures.
                                                                                                     v 5.40
1    Introduction

The organization of markets has long posed a rich set of questions to economists. Since the
seminal experiments of Smith (1962, 1964), economists have realized that the institutions
used to aggregate information and determine prices in a market can have substantial conse-
quences for the market’s efficiency. In recent years, this line of inquiry has turned normative
as a new field of market design has emerged (Roth, 2002). It seeks to identify specific market
rules and procedures that can speed information aggregation, discover prices, and improve
efficiency relative to less-structured trading arrangements.

    This paper examines how market organization affects performance, efficiency, and prices
in competitive electricity markets. In most regions of the United States, wholesale electricity
markets operate as a decentralized, bilateral trading system. In a handful of areas, however,
trade is mediated through centralized market designs. These markets aggregate offers to buy
and sell, determine market clearing prices, and handle settlements for several complementary
services involved in energy production and its delivery.

    During the past decade, this heterogeneity in market organization has received increasing
attention in the policy arena. In 2001, the Federal Energy Regulatory Commission (FERC)
initiated a formal proceeding to identify ‘best practices’ in electricity market designs, and
to promote their use in regions where bilateral trading practices prevail. The FERC’s policy
initiative has been vigorously challenged by market participants who expect to lose from
a change in market outcomes. Their central challenge is that the benefit of expanding
organized market designs into new regions of the nation remains speculative, and may not
be worth the cost of implementation.

    The fundamental premise of the FERC’s initiative is that adopting a well-chosen mar-
ket design would improve market efficiency in areas where decentralized, bilateral trading
practices prevail. This paper provides direct empirical evidence on this hypothesis. We
do so by examining an abrupt change in a regional energy market’s organization. In 2004
the largest organized electricity market in the United States, known as the PJM Intercon-
nection, expanded to include nineteen firms that formerly traded electricity exclusively in
Midwestern bilateral markets. The new members include the largest power producer in
the nation, the American Electric Power Corporation, which shifted an enormous volume
of trade onto PJM’s market. For technical reasons, this change in the venue of trade was
implemented on a single day. That enables us to evaluate the impact of change in the
market’s organization, from decentralized bilateral trading to a centralized auction market
design, in a field setting of considerable economic consequence.

   The questions we pursue can be examined in laboratory market experiments, and an
impressive body of studies have attempted to do so (see especially Olson et al., 2003).
Nevertheless, persuasive empirical evidence with respect to electricity market designs has
remained elusive. In general, the efficiency of unstructured bilateral markets depends how
buyers and sellers are matched to one another. Without strong assumptions about how
this matching occurs and market participants’ information sets, theory (and laboratory ex-
periments) admit a wide range of efficiency outcomes with decentralized trade. In real-life
decentralized markets, participants’ information sets are difficult for researchers to observe
and characterize, making the relative efficiency of decentralized versus organized markets
difficult to establish outside the field. To make informed decisions, it is desirable to deter-
mine empirically whether the efficiency gains of an organized market design are realized,
and compare this magnitude with the cost of implementing it.

   This objective differs from much of the burgeoning theoretical and empirical literature
on electricity markets. Previous empirical contributions have focused, more or less exclu-
sively, on whether organized electricity markets operate efficiently relative to a ‘perfectly
competitive’ market benchmark (Mansur 2007, Borenstein Bushnell and Wolak 2002, Wol-
fram 1998). In contrast, our objective is (1) to provide evidence on whether the widely-used
bilateral market arrangement operates efficiently or not, and if not, (2) whether shifting the
venue of trade onto an organized market improves market efficiency. In other words, we
are after practical comparisons regarding relative efficiency of the two workable, existing
trading systems we observe in use. This seems the salient market-institution comparison to
inform standing market organization debates.

   Our empirical strategy exploits several attractive features of our setting. First, we have
both market-level and plant-level data on supply, demand, prices, quantities, and produc-
tion costs. This detailed information allows us to assess how market organization affects
market performance along several related dimensions, as well as to construct measures of
the economic efficiency gains. The efficiency gains ultimately arise from supply-side al-
locative efficiency improvements, as trade reallocates production from higher-cost plants to
lower-cost plants.

   Second, the synchronized market change date yields a clear ‘before versus after’ demar-
cation of the two forms of market organization. There were no concurrent changes in the
number or the identities of the firms operating in these markets, nor their technologies or
capacities. Nor were there any changes in the physical infrastructure of the (transmission)
network that enables them to trade. Instead, what changes is how the market partici-
pants’ willingness to buy and sell was elicited, aggregated, and used to determine market

prices. The organized market we examine uses this information to price a richer set of
state-contingent commodities, and it provides price transparency that did not exist with
decentralized, bilateral trading. We find that this enables the organized market to direct
production to the most efficient available resources, realizing greater gains from trade than
occurred under the bilateral trading system.

    We describe the organization of the markets we study in more detail in the next section.
We present in Section 3 our empirical strategy for identifying efficiency changes on the
basis of observed market outcomes. Section 4 describes the data, and the changes in prices
following the market’s reorganization. Evidence on corresponding real production changes
is presented in Section 5. We then present a model of these price changes that enables
us to estimate the magnitude of the efficiency gains after the new market organization’s
implementation. A discussion of these findings and their implications concludes.

2     The Markets

One of the most controversial issues surrounding the liberalization and restructuring of
the electric power industry is the design of wholesale electricity markets. It is useful to
summarize why this uncertainty persists, relating the economic issues to the specific markets
we study.

2.1   Background

As noted above, regional electricity markets in the United States use one of two different
organizational forms: decentralized bilateral trading or organized (auction-based) markets.
We use the term decentralized to characterize markets in which the distinct services needed
to trade electrical power (generation, transmission, and ancillary support services) are ob-
tained in separate, sequentially-arranged transactions. In practice these services are pieced
together bilaterally, with different counterparties supplying production (generation) and
delivery (transmission and support) services.

    The theoretical appeal of organized market designs for electricity arises because there
are strong complementarities between these distinct services. These complementarities exist
on both sides of the market. On the demand side, the price a buyer is willing to pay for
generation depends acutely upon the availability and price of transmission services. On
the supply side, the price at which seller will provide various network support services

required for delivery depends upon the market price of energy (generation). Well-crafted
market designs set a clearing price for each of these complementary services simultaneously,
seeking to mimic an efficient (Walrasian) market equilibrium.

    In principle, decentralized markets can achieve an efficient allocation of complementary
services through sequential trade in each constituent commodity—if market participants
have sufficiently rich opportunities to contract and re-contract. The market price of each
service should continually readjust after each transaction until the market’s allocation (of
all services) admits no further gains from trade. In practice, various technical features
of electricity markets severely restrict these adjustment opportunities; efficient market-
clearing prices vary over space and time, and often change rapidly. It remains an open
question how well the separation of markets for key complementary services in decentralized,
bilateral electricity trading actually discovers these prices. Unfortunately, this concern is not
purely theoretical; rationing of network (transmission) service between locations is routinely
observed in areas that employ bilateral trading, suggesting that the decentralized markets
do not always match supply and demand efficiently—or even clear the market.1

    In reality, what matters most is not whether one form of market organization or the
other achieves a theoretically ideal market outcome, but whether the difference between
them is economically significant. Industry participants in regions of the U.S. where bi-
lateral trading prevails commonly argue that the cost of adopting (or joining an existing)
organized market would exceed the benefit. At its core, this points to a central trade-off
in market design: An organized market design might reduce inefficiencies that exist in an
unstructured, decentralized market, allowing participants to realize gains from trade that
would not otherwise be achieved. However, organized markets are costly to design and
implement—particularly so for electricity. The value of shifting the venue of trade out of a
decentralized bilateral system and into an organized market is an empirical question.

    This trade-off has emerged as a policy issue for two reasons. The liberalization of
wholesale electricity markets since the mid-1990s allows (essentially all) buyers and sellers
to trade at market-determined prices. The role of the industry’s regulator (the FERC)
is now effectively one of market oversight.2 However, the FERC retains an obligation to
evaluate and approve changes in electricity market designs—a task that is not taken lightly
in the wake of California’s market disaster.
      Rationing of transmission service does not imply that total energy demand exceeds total production; its
consequence is allocative inefficiency, leading a higher-cost producer to operate while a lower-cost producer
elsewhere in the network does not.
      This is not true of retail electricity markets, the regulation of which varies state by state.

   The difficulty that policy makers face is that their goal of encouraging more efficient
markets is not always aligned with the private incentives of market participants. For in-
stance, a producer may have a strong private incentive to object to a new market design if it
will result in a more competitive marketplace with lower prices; and a buyer that relies upon
a particular (transmission network) path for delivery may object to a market change that
would price that delivery path based on its highest-value use. The practical consequence of
these fundamental incentive compatibility problems is that regulatory policy makers face a
panoply of conflicting claims about the costs and benefits of organized market designs.

   Given these circumstances, it seems particularly useful to examine how market organi-
zation affects market efficiency. We bring new information to this problem by examining
how market outcomes changed after an existing, organized electricity market expanded to
serve a region where an (exclusively) bilateral trading system prevailed. This seems an ap-
propriate setting to examine empirically in order to inform electricity market organization
decisions. The abrupt timing of the change we examine, and the rich market data available,
will enable us to draw useful inferences about the relative efficiency of bilateral trading
versus this organized market design. We also examine precisely why this change in mar-
ket organization affected overall market efficiency, pointing to informational problems that
well-crafted market designs can overcome but which present severe difficulties for bilateral
trading arrangements.

   From a research standpoint, a key issue is how we can use the market outcomes we
observe to inform the relative efficiency of these two different trading systems. This requires
some attention, which we provide presently. First, a brief summary of the specific markets
we will examine.

2.2   The PJM Market Expansion

The organized market we examine is the PJM Interconnection (PJM). PJM operates several
inter-related wholesale markets for electricity (energy), its delivery, and a variety of ancillary
services. The four-hundred members of PJM are primarily local utilities that buy electricity
to distribute to homes and businesses, producers that own power plants, and third-party
traders in PJM’s forward markets. PJM operates spot (hourly) and forward markets for
electricity production and delivery at thousands of delivery points (nodes) across a broad
region of the U.S., ranging from New Jersey to Virginia and west to Illinois. The nominal
value of all transactions on PJM’s spot and forward markets annually exceeds $22 billion.

    In contrast, utilities and power producers throughout most other regions of the Eastern
U.S. (excepting the Northeast) engage in wholesale electricity trading through bilaterally-
negotiated transactions. Until the PJM expansion in 2004, this bifurcated market organi-
zation persisted, in a more-or-less stable form, for more than a decade. Following several
years of planning and regulatory approvals, in October 2004 nineteen Midwest-based firms
that previously traded exclusively through bilateral market arrangements became members
of PJM. Seven of these new members are affiliated subsidiaries of the American Electric
Power Company (AEP), a holding company that, until joining PJM, was one of the largest
participants in regional bilateral markets in the Midwest. The new members increased the
total volume of energy priced on PJM’s markets by approximately $3.5 billion annually.

    The decision of the new exchange members to join PJM originates in an (unrelated)
merger settlement with federal authorities half a decade earlier. Whether that decision
reflects accurate forward-looking behavior by AEP and others (about the consequences of
joining the organized market) is an interesting question, and one that affects how we will
interpret the results. It does not, however, affect our ability to identify whether PJM’s
expansion improved market efficiency overall. We discuss this issue further below, after a
more detailed discussion of how we assess efficiency changes in this setting.

3    Identification: Market Efficiency and Market Organization

In our setting, identifying how the organized market’s expansion affected market efficiency
entails three related, but conceptually distinct issues. The first issue is how we use the mar-
ket outcomes we observe—prices and quantities, primarily—before and after the market’s
expansion to infer changes in market efficiency overall. This we describe next.

    The second issue is the question of cause and effect: Whether, and why, we may be
confident that any changes in market efficiency we measure are attributable to the markets’
expansion, and would not have occurred otherwise. This stems from the timing and nature
of the changes we study.

    The third issue is an understanding of the causal mechanisms: From a microeconomic
standpoint, what underlying market mechanisms explain why this organized market design
improves market performance, relative to the bilateral trading system? This issue takes us
into the differences in trading arrangements of the two trading systems. We take this up
further below.

3.1      A Conceptual Framework

Since electricity markets are complex in their fine details, we begin with a simple market
analogy. This analogy conveys the essential market features we exploit to identify efficiency
changes using observable market data.

   Imagine a market with many participants who have heterogeneous, privately-known
valuations. Participants trade with one another bilaterally, at prices determined in private
negotiations. Suppose further that some of the market’s participants are also members of
an exchange, or clearinghouse, that matches offers to buy or sell among its members in
an organized fashion. (The precise details of this process are not yet required). Exchange
membership is open to any participant who pays a (fixed) membership fee. The exchange
members are free to transact with non-members, but must do so outside the exchange in
the bilateral market.

   To complete the analogy, imagine now that a subset of the bilateral-market partici-
pants joins the organized exchange. Following our earlier terminology, we will refer to the
two transaction venues in our simple analogy as the bilateral market and the organized
(exchange-based) market.

   In this setting, the expansion of the exchange’s membership could have several possible
consequences for the markets’ performance. In the absence of any trading frictions in the
bilateral market, we expect the exchange’s expansion to have no effect on the markets’
transaction prices nor the aggregate quantity traded. The lack of arbitrage impediments in
the bilateral market, where all trade between exchange members and non-members takes
place, implies that bilateral- and exchange-based transactions should occur at the same

   Empirically this turns out not to be the case, so an alternative maintained hypothesis
about trading frictions is needed. Suppose now that contractual incompleteness, search
costs, or some other trading imperfection exists in the bilateral market. In this case we ex-
pect a non-zero price spread between the bilateral and organized markets, and an incentive
for some market participants to join the exchange (depending, given traders’ heterogeneous
valuations, upon the opportunity cost this spread represents relative to the cost of member-
ship). In our application, we will draw conclusions about relative efficiency of the bilateral
and organized markets. These emerge not from the fact that some market participants
joined the organized market per se, but instead from changes in the two markets’ prices
and quantities after they joined it.

   At one level, the logic underlying our empirical strategy is straightforward. In this simple
analogy-and in reality-any pair of market participants has the option to transact bilaterally
outside the organized market. But the exchange has a membership cost. Thus, if we observe
an increase in the quantities transacted by the new exchange members after they join the
organized market (ceteris paribus), we conclude that the new market participants realized
gains from trade that they could not capture by transacting in the bilateral market.

   This logic carries over to inference about market efficiency on the basis of price changes,
although the argument is slightly more subtle. By itself, a non-zero price spread provides
little (if any) information about the relative efficiency of the bilateral versus the organized
market. A spread simply indicates there exist barriers to trade in the bilateral market,
since that is where all transactions between exchange members and non-members must
take place. What does provide useful information about market efficiency improvements is
a change in this price spread.

   The logic for why an expansion of the organized market might alter this spread is that it
changes the distribution of valuations among the participants in each market. For instance,
suppose (without loss of generality) that prices in the bilateral market are lower than in
the organized market. Then sellers have an incentive to join the exchange, withdrawing
(or raising the offered price for) supply in the bilateral transaction market and expanding
aggregate supply in the organized market. Such a shift narrows the price spread between
the bilateral and the organized market, increasing the volume of trade overall.

   In sum, after the new exchange members join the organized market, efficiency-enhancing
reallocations from low- to high-value market participants will reduce the (magnitude of the)
price spread between bilateral- and exchange-based transactions. Thus, a central component
of our empirical strategy will be to evaluate whether price spreads converged significantly
after the organized market’s expansion.

   The subtlety in this otherwise straightforward microeconomic argument is that it rests
on imperfect trade within the bilateral market. The logic is immediate: If not for such im-
perfections, then any opportunities for trade between exchange members and non-members
would be arbitraged away in the bilateral market. If that were the case, we would see no
changes in observable market outcomes after the organized market’s expansion.

   The importance of the imperfections in the bilateral market to our efficiency analysis
is the following. The economic arguments above allow us to support empirical conclusions
about whether or not market efficiency improves after the organized market expands, using
observable changes in price spreads and quantities transacted. The magnitude of these

efficiency gains informs the opportunity cost of employing a decentralized, bilateral trading
system. That is the central empirical question we seek to inform.

    Still, the market price and quantity evidence, by itself, admits a number of underlying
causal mechanisms for imperfections in a decentralized bilateral trading system. These
include the complementarities in electricity markets that we discussed at the outset of
Section 2, or that there are search costs when locating counterparties with mutual gains
from trade outside an organized exchange venue, and a few other mechanisms. Resolving
among these underlying economic explanations requires additional information about how
bilateral market trading works.

    First, we discuss the empirical results.

4     Price Spreads

We now examine whether the price spread between comparable transactions priced in the
bilateral market and in the organized market (PJM) changed after PJM’s expansion. Be-
cause the details of how prices are measured are important to our purposes, we first discuss
the underlying data.

4.1    Price Data

To examine whether between-market arbitrage improved, we assembled detailed market
price data, at daily frequency, for a two-year span. There are two data sources for transac-
tion prices in bilateral electricity markets, the Platt’s daily price survey and the electronic
‘over the counter’ trading system operated by the Intercontinental Exchange, Inc. We have
examined daily transaction data from both sources, and the daily price indices for the de-
livery points of interest are (essentially) identical. In the results below we have used the
Platt’s data due to its slightly broader coverage, unless indicated otherwise. The prices
determined by PJM are public information, which we obtained directly from PJM.

    Because electricity must be produced at precisely the moment it is used by consumers,
trading in wholesale electricity markets is primarily conducted on a forward basis. Our
analysis centers on prices in the day-ahead forward markets. Day-ahead forwards are the
highest-volume markets for wholesale electricity transactions, in both the bilateral and the
organized market.

       In principle, the bilateral market and exchange-based (PJM) day-ahead forward prices
we compare represent identical commodities, up to their delivery point. Each indicates the
price for delivery of the identical quantity of power, at the specified delivery location, for
a pre-specified duration the following day. In bilateral markets, two standard contracts are
traded: Peak and off-peak, in 50MW units, for next day delivery continuously from 7am to
11pm or 11pm to 7am (EPT). On PJM, separate prices are set for each hour of delivery;
we construct the equivalent prices for the industry-standard peak and off-peak delivery
intervals, thereby matching exactly the delivery schedules for the contracts traded in the
bilateral market.

       These contracts differ in one respect: PJM’s day-ahead markets use different pricing
(delivery) points than bilateral market forward contracts. This will affect our analysis and
interpretation, as discussed below. In terms of the data, we selected a set of delivery points
in the mid-Atlantic and Midwestern states that are most likely to reveal changes in market
outcomes as result of PJM’s expansion (into the Midwest).3 Formally, to select a set of pric-
ing (delivery) points, we used three simple criteria: (1) Proximity of each delivery point to
one another (where proximity is with respect to structure of electric transmission network);
(2) commonly-used delivery points, to insure liquidity; and (3) for which comprehensive,
location-specific day-ahead market price data exist. There are exactly five delivery points
that meet these criteria in use. Rather than select among them, we will report results for all
five points and the price spreads between them. All of our results and their interpretations
turn out to be highly robust to the choice of which delivery points to compare between PJM
and the Midwestern bilateral markets, as will become clear presently.

       There is a second, minor difference in the pricing of day-ahead forward contracts, due
to the timing of each market’s close. Bids in the PJM forward market are due by noon on
the day prior to delivery, at which point the day-ahead market closes. Prices are posted by
the market by 4pm. Bilateral market price data include trades up to close of business day.
Thus the information set of traders in bilateral markets is a superset of that incorporated
into the organized market’s day-ahead prices. Nevertheless, there is no reason why any
additional information would bias bilateral market prices one way or another, relative to
PJM’s day-ahead forward prices.
    In this respect our analysis is clearly a partial, rather than general, equilibrium analysis of the expansion’s
impacts. We have not included analysis of additional delivery point prices here primarily to reduce the volume
of our analysis—other, more distant pricing points might also be affected by expansion, ostensibly by lesser

4.2      Pre- v. Post-Expansion Price Spread Changes

Table 1 summarizes the price levels and price spreads between the bilateral market and the
organized market before and after the market’s expansion on October 1, 2004. The first
column of data presents average daily forward prices by delivery point over the six-month
period prior to expansion, and the average (absolute) price spread between the two markets’
delivery point pairs.

   As the table indicates, average prices tend to differ at different delivery points, an
empirical regularity in electricity markets generally. The standard explanation for these
price differences is that they reflect occasional congestion on the transmission network used
for delivery. That is, when the difference in prices between any two delivery points creates
excess demand for delivery (transmission capacity) from one point to the other, the market
may not be able to close the price spread completely. In essence, arbitrage in these markets
is limited by the network’s capacity constraints, which can be binding at times. In an
efficient market, the price spreads would be zero when there is excess capacity and positive
(in magnitude) if not; positive price spreads we see in average prices reflects a mixture of
these two conditions that varies day to day.

   The presence of non-zero price spreads due to network congestion between delivery
points has an important implication for our analysis. We are not interested per se in testing
whether arbitrage is ‘perfect’, in the sense of continuously equating prices between these
market-specific delivery points. Rather, we are interested in assessing whether arbitrage
improves as a result of the market’s expansion. In other terms, the central question here is
whether markets find better ways to use the existing network capacity in place to increase
trade, thereby reducing price spreads.

   The second column of data shows the comparable average prices and price spreads for
the six months post-expansion. After PJM’s expansion, the organized market now sets
a price for delivery at the same central Ohio hub (AEP-Dayton) as the bilateral market.
For comparability, the prices shown in this column for all three bilateral market delivery
points correspond to actual bilateral market transactions. The PJM market price for the
AEP-Dayton hub is the same as that reported here for bilateral-market transactions at that

   Price Spread Convergence. Changes in prices and in the price spreads between
markets are listed in the third column. These indicate that the price spreads changed at all
locations from pre-expansion levels, and in a striking way. For the peak-period contracts

in Panel A, price spreads converge at all six bilateral-PJM delivery point contrasting pairs.
The magnitudes are similar across all six contrasts, ranging from −$1.53 to −$3.05 per
megawatt hour. In percentage terms, the decline in the absolute price spreads ranges from
34 to 41 percent of the pre-expansion average price spread.

    The change in the price spreads between market is even more dramatic for the off-peak
delivery period, shown in Panel B. The bottom portion of this panel shows the changes
in the off-peak price spread for all six bilateral-PJM delivery point contrasts. Again, the
absolute price spreads between the two markets fall by similar magnitudes for all six pairs,
ranging from −$5.78 to −$7.73 per megawatt hour. These correspond to 60-to-75 per-
centage point declines from average pre-expansion absolute price spreads. The changes in
the absolute price spreads between markets are also large relative to the variance in daily
spreads. The convergence of all of these price spreads, peak and off peak, are highly sta-
tistically significant, using nonparametric (Newey-West) autocovariance-adjusted standard

    Interpretation. Table 1 shows that bilateral-exchange market spreads fell substantially
post-expansion. The formal change here is that a set of market participants, who previously
traded electric power in exclusively in bilateral markets, joined PJM’s organized exchange.
In doing so, PJM began pricing energy and transmission service in a portion of the Midwest
proximate to (or including) the bilateral-market delivery points listed in Table 1.

    This does not imply, by any means, that the firms that joined the exchange did not
continue to participate in the bilateral market. Indeed, the simplest interpretation of the
main findings in Table 1 is that the new members of the organized market found addi-
tional trading opportunities through it—with counterparties who were existing members
of PJM. Any existing exchange member that previously bought at PJM’s Western Hub or
Allegheny delivery points faced systematically higher prices than were available where the
new exchange members produce, near the bilateral-market delivery points. After the new
exchange members joined the organized market, the PJM market design rapidly identified
these trading opportunities and increased the new PJM members’ total energy production.
Since electricity producers’ marginal costs increase with output (at the firm level), expanded
production increased the new exchange members’ costs—and therefore the price at which
the new PJM members are willing to sell to their previous trading partners in the bilateral
     There is evidence of slight persistence in the price spreads between delivery points. This is likely
attributable to the fact that exogenous changes in network capacity that create congestion (weather distur-
bances and line deratings) may last more than one day.

       Alternative Time Horizons. In Table 1, we present price information based on a
six month ‘window’ pre- and post-expansion. This relatively long horizon is used here
because the economic importance of any change in market outcomes depends upon whether
it persists over time.5 We have also replicated the analysis in Table 1 using pre- versus post-
expansion ‘windows’ of 30 days and 90 days. These yield quantitatively similar changes to
those shown in Table 1, in all six between-market price spread contrasts, for both the peak
and off peak contracts. (The shorter window tables are available from the authors).

       More remarkably, these data also indicate that average spreads fell quite quickly after
the new members joined PJM. For example, the peak-period price spread between the
PJM Western Hub and the Cinergy delivery points, which are the most liquid contracts in
each the two markets, was $14.09 the day before market expansion (October 1, 2004); the
price spread one day after integration was $9.37, a decline of 33 percent. Comparing the
average (absolute) spread for ten days before and after October 1, 2004, we see a similar
change of −$3.61 per megawatt hour, a decline of 42 percent. Regardless of the window
length examined—from one day out to six months after the market expansion—we see peak
period price spreads fall by 30-40 percent from pre-expansion levels, and slightly more than
double this amount for off-peak price spreads. In sum, the results reported in Table 1
appear highly robust to the choice of window length. The price spreads between markets
fell quickly after PJM expanded, and remained far smaller thereafter.

       Did same thing happen in prior years? The answer is no, unequivocally. A simple
‘placebo analysis’ is to replicate these calculations for a comparison periods centered on
October 1, 2003. There we see no changes in average price spreads, whether we evaluate
them with window lengths of six, three, or one month or one week. We omit the detailed
tabulations here, as we will show this quantitatively using a more sophisticated comparison-
year statistical test in Section 6, below.

       Before we turn to a more formal model of price spread convergence, however, it is useful
to examine some quantity data. These data help paint a more complete picture of how the
market’s organization change altered trade.
    In May 2005 a new set of players joined the PJM market, dramatically expanding it once again. We
limited our post-expansion attention to six months here solely to avoid confounding these results with further
price spread changes post-May 2005.

5        Quantity Evidence

If the convergence in prices reflects true gains from trade, then it should be the result of
changes in the quantities traded between market participants. We examine this first using
data on quantities of power transferred between delivery point areas.

5.1       Quantity Changes from Aggregate Transfers Data

The changes in the price spreads shown in Table 1 are sufficiently large that they drew
considerable attention from energy traders and producers. Although electricity trading is a
fairly specialized business, the Wall Street Journal ran a front-section article shortly after
the PJM expansion on the dramatic changes in power flows and prices in this area of the
U.S. One quantitative piece of information presented in it came from the Vice President for
Operations at American Electric Power (AEP), the largest (by far) of the firms that joined
PJM, who reported that AEP’s shipments of power eastbound to other PJM members
tripled after the market’s expansion.

        To confirm this, we obtained information on the scheduled and actual (real-time) power
flows between these parties, before and after the market’s expansion. The interpretation of
power flow information is most transparent by expressing the aggregate quantities trans-
ferred as a share of the network’s transfer capacity between them. Figure 1 shows the
day-ahead scheduled quantity of power to be transferred between the AEP-Dayton delivery
region and the PJM delivery areas listed in Table 1, from October 2003 through mid-2005.
The market expansion date is indicated by the dashed vertical line at October 1, 2004.6

        The striking feature of Figure 1 is the dramatic increase in the quantity of power shipped
between these two areas the day after the market’s expansion. Prior to that date, AEP was
not a member of PJM’s organized market. It could (and, according to our data, did) trade
with members of PJM during this period, but all trading between them was necessarily
conducted in the bilateral transaction market. After October 1, 2004, AEP is a member
of PJM; now the PJM market design identifies trading opportunities between all of its
members. Since AEP’s production facilities are based in a lower-cost region of the midwest
than PJM’s members to the east, the efficient allocation of production entails maximizing
the quantity transferred between them (up to the point of equalizing prices, if feasible).
    We have calculated the flow information summarized in Figure 1 using net flows, where a westbound flow
would be indicated by a negative utilization rate in Figure 1. There are very few hours in which transfers
are ever westbound.

When trade between AEP and the pre-existing PJM members shifts out of the bilateral
market and onto the PJM market design, the result is that it maximizes the quantities
produced and traded between AEP and other PJM members—right up to the limits of the
transmission networks’ capacity.

       Two technical aspects of Figure 1 merit note here. First, the underlying data we have
used to construct this figure are hourly; for transparency, here we have aggregated the hourly
data to weekly average transfer capacity utilization rates. The economic interpretation is
unaffected. Second, the entities that record and monitor these flows changed their data
conventions with the market’s expansion; the results before the expansion are based on load
flow data (in megawatts) and first-contingency total transfer capability between regions,
while the post-expansion data are derived from day-ahead congestion frequency data. The
consequence of this is that the pre-expansion transfer capacity utilization data in Figure 1
should be interpreted as an upper bound on the flows during this period. We discuss this
in more detail in the Appendix.

       Second, the data depicted in Figure 1 are the day-ahead scheduled power flows that
correspond to the day-ahead forward market prices summarized in Table 1. There are also
balancing market prices (at PJM delivery points only, however), and real-time engineering
power flow data between these regions. Like Figure 1, the real-time power flow data show a
dramatic increase in quantities transferred (eastbound) between the AEP-Dayton and the
PJM Western Hub/Allegheny delivery regions on October 1, 2004. However, the average
capacity utilization rate over the post-expansion period in the real-time data is lower, at
76% (as opposed to 96% in Figure 1). This difference is difficult to explain, although
we speculate that it may result from transmission capacity ‘hoarding’ by network system
operators who determine and provide capacity information to the day-ahead market.

       An additional piece of corroborating data comes from an independent study of power
transfers in this region during 2004 (EISA, 2005). The authors used engineering data and
models to evaluate changes in the quantity of power transferred across the transmission
network in this region. The EISA study is notable because the authors had access to
actual, telemetered power flow data across specific power lines in the transmission network
between the AEP-Dayton delivery region and the PJM Western Hub region. Their results
are similar to the evidence shown in Figure 1, indicating a dramatic net increase in (west
to east) average hourly energy transfers between September and October 2004.7
    The EISA author’s report a smaller total quantity change (750 megawatts per hour, on average), because
the lines they examine comprise only a subset of the total transmission transfer capacity between these two

   In sum, it is clear that there was a dramatic increase in the flow of power eastbound,
across the interface that separates the bilateral-market delivery points and the organized-
market delivery points summarized in Tables 1 and 2. This is economically equivalent to a
substantial increase in arbitrage in the day-ahead forward markets. In practice, the change
in trading arrangements accompanying the organized market’s expansion imply this arbi-
trage is taking place anonymously, through PJM’s organized day-ahead forwards market.
Effectively, the new members joining PJM were matched by the organized market system
to new buyers elsewhere in PJM (to the east), increasing the new members production from
generating assets physically located near the AEP-Dayton delivery region and decreasing
production from other generating assets located closer to the PJM Western Hub and PJM
Allegheny delivery regions (or still further east). The result of these quantity changes is the
power flow transfer changes between these points summarized in Figure 1.

   To put these magnitudes into perspective, an interpretation may help. If, to be con-
servative, we take the actual (real-time) quantity data on power flows, the increase we see
between the average pre-expansion and post-expansion transmission utilization rates here
corresponds to approximately 1,300 megawatts of power per hour. In this region of the
United States, this is enough power to simultaneously serve well over a million households.
The changes in where power is being produced are large indeed.

5.2   Quantity Changes in Plant-level Production Data


   At a microeconomic level, there are efficiency gains from trade in wholesale electricity
markets when trade reallocates production from high-cost plants to lower-cost plants. While
the aggregate transfers results in the previous subsection indicate a major quantity reallo-
cation took place post-integration, it is useful to examine whether there is corresponding
evidence for changes in production at the individual plant level.

   To do this, we turn to a new, additional data source. To track compliance with air
emissions regulations, the US Environmental Protection Agency continuously monitors the
operating performance of major fossil-fueled power plant in the United States. Their Contin-
uous Emissions Monitoring System (CEMS) database contains the hourly fuel consumption,
emissions production, and electricity output for most power plants in our markets, nuclear
plants excepted. We use the CEMS data to assess changes in output after the organized
market’s expansion, at the individual generating-facility level.

   There were more than 600 generating units owned by members of the organized market
at the time of its expansion. However, the natural place to look for changes in production is
among the subset of plants whose owners joined the organized market on October 1, 2004.
These are the plants likely to have realized new opportunities for trade under the organized
market’s design. There are 66 such generating units in the CEMS data.

   Two findings stand out. First, the large changes in price spreads during the off-peak
period, as shown in Panel B of Table 1, suggest we should see quantity changes for the
largest, relatively efficient coal-fired power plants that are typically on the margin during off-
peak periods. In fact, in our 66-unit group the six largest generating units saw a 22% increase
in off-peak production over the six months post-integration relative to the six months pre-
expansion. This is an enormous output increase, equal to approximately 1100 megawatts
per hour (on an average hourly basis). By comparison, over the same calendar-month
periods one year earlier these six plants’ total off-peak production was virtually unchanged,
at −0.2%. All six of these generating units are physically located in the area indexed by the
AEP-Dayton market pricing point. We are led to conclude that the substantial reduction
in off-peak price spreads in Table 1 accompanied a large increase in quantities traded, and
much of that came from these six, low-cost power plants.

   The second finding concerns the total change in production for these 66 units as a
whole. Normally, aggregate electricity production in this region of the U.S. is lower in
the six months spanning winter (October through March) than the six months spanning
summer. In the data for 2003, the total output for all 66 production units decreased by 140
megawatts (on an hourly average basis) from the six-month summer period to the following
six-month winter period. In sharp contrast, the total output of all these units increased
by 800 megawatts (on an hourly average basis) between the six-month period preceding
the market’s expansion and the six months post-expansion (October 2004). A change in
seasonal production of this magnitude by such a large portfolio of generating stations is
difficult to explain by any means other than that the owners of these plants found new
opportunities to sell to distant buyers starting in October 2004.

   This net increase in production has an important additional implication for interpreting
how improved efficiency arises post-expansion. The buyers and sellers in these markets are
interconnected to a large number of other regional power markets throughout the Eastern
U.S. As a result, the increase in quantities shipped eastward across the (transmission)
network described in Section 5.1 could, in principle, reflect two different forms of arbitrage.
One is an increase in actual production by a low-cost seller that joined the organized market,
displacing higher-cost production elsewhere. Alternatively, the changes in (transmission)

networks flows could reflect a shift in the destination market for the output from another,
more distant low-cost seller who did not join the organized market at all.

        The distinction between these two possibilities relates to our understanding of how the
organized market’s expansion improved efficiency. Recall that since only members of PJM
can trade in its forward markets, prior to the market’s expansion the exchange’s initial
members and its members-to-be transacted only in the bilateral market. Our conjecture is
that the organized market is a more efficient market venue than this decentralized bilateral
market, resulting in increased gains from trade post-expansion. However, if the changes
in trading patterns observed in the network flow data in Section 5.1 were the result of a
production reallocation between exchange members and producers that did not join PJM
when it expanded, then our conjecture would require some amendment.

        For this reason, it is reassuring to observe in actual, metered-at-the-plant generation
data that the price spread convergence matches large-scale production changes by the firms
that joined the organized market. These are the firms and power plants that must have
changed behavior if the theory that overall market efficiency improved after they joined the
market is to find support. Here we find that the increase in quantities transferred between
delivery areas (in Figure 1) is due primarily to an increase in physical production by sellers
who joined the organized market, and whose plants previously operated at lower levels
before the market’s expansion.8

        The magnitude of the quantity changes and price spread declines that followed the
organized market’s expansion suggest that the gains from increased trade are substantial.
Our next task is to refine this analysis and provide quantitative evidence on the magnitude
of the economic efficiency gains.

6        An Empirical Model of Imperfect Trade

We now present an empirical model of imperfect trade that is suited to our setting. Our
primary purpose here is to be able to estimate the economic significance of the changes
in price spreads that occurred after the organized market’s expansion. This necessitates
information on the elasticity of supply, and how it varies pre- and post-expansion, for
the various delivery points affected. The model we present and estimate provides this
    One useful, if anecdotal, piece of additional corroborating evidence comes from AEP’s 2005 Annual
Report: It describes the increase in output and profitability of this firm’s low-cost, high-capacity coal
generating capacity after AEP joined the PJM market (AEP, 2006).


   In addition, we have a second objective. Over a very short (one day to one week)
horizon, it is difficult to conceive that actual plant-level marginal costs in these markets
change appreciably. However, over longer horizons this assumption becomes suspect. In
particular, during the winter of 2004, the price of the fossil fuels and emissions permits that
are the primary variable factors of production for power producers rose significantly. In an
environment were network congestion occasionally serves as a barrier to trade, it becomes
conceivable that changes in these factor prices could account for a portion of the changes
in price spreads over time. A similar argument can be made with respect to weather and
factors that vary consumer electricity use from day-to-day or season to season.

   Consequently, before we base economic efficiency conclusions on the observed changes in
price spreads over longer horizons, it is desirable to determine quantitatively if any portion
of these changes can be attributable to exogenous changes in factor prices or retail electricity

6.1   The Model

To model the effect of integration on price spreads, it is useful to imagine a simple two-
sector trade model with imperfect trade between them. Here the two sectors correspond
to the organized market and the bilateral market. Consider Figure 2. Let S1 (Q) be the
inverse supply function of all producers who are members of the organized market, and let
Qd be the consumption of final consumers directly served by the members of the organized
market (e.g., local distribution utilities). Because retail electricity prices are regulated and
change (typically) only on an annual basis, electricity consumption is insensitive to day-to-
day changes in wholesale electricity market prices. Consequently, aggregate consumption is
indicated by the dashed vertical line.

   Similarly, let S2 (Q) be the inverse supply of all producers who are not (initially) members
of the organized market. Here Qd represents the electricity consumption of final consumers
served by firms that are not part of the organized market. In Figure 2, we have reversed
the horizontal axis for market 2, so S2 increases to the left.

   In the absence of any trade between the organized market’s members and non-members,
the prices that would prevail in each market are indicated by p∗ and p∗ . We term this the
                                                               1      2
autarky outcome. Since there is always some trade between them, however, we do not ob-
serve p∗ and p∗ directly. Instead, we observe prices p1 and p2 in each market, corresponding
       1      2

to a quantity of trade equal to (the width of) the shaded region in Figure 1. Note the prices
that we observe in each market (p1 , p2 ) are not assumed to capture all gains from trade,
although they admit this possibility.

   In this two-market model, the effects of improved arbitrage are manifest through changes
in quantities supplied, not through shifts in the supply curves themselves, nor through de-
mand changes. By contrast, changes in factor prices or (putatively exogenous) retail elec-
tricity consumption (e.g., due to weather) will shift supply or shift demand. Distinguishing
these potential sources of price spread variation from improvements in arbitrage activity
per se requires being able to separate out shifts of supply from movement along supply.

   To do this, we relate the observed price spread to the supply and demand fundamentals
in each market. Since some trade occurs both pre and post, observed price spreads are
bounded by the autarky spreads: |p1t − p2t | ≤ |p∗ − p∗ |. Equivalently, think of arbitrage
                                                 1t   2t
as reducing the markets’ autarky price spread by a (random) proportion νt each period:

                                   |p1t − p2t | = νt |p∗ − p∗ |
                                                       1t   2t                            (1)

where νt ∈ [0, 1]. At the extremes, νt = 1 implies no gains from trade are being realized
and νt = 0 implies an efficient, integrated market. The exact (marginal) distribution of νt
is unrestricted.

   Changes in exogenous market conditions that shift supply and demand—factor prices,
weather conditions, or the like—alter observed prices by changing p∗ and p∗ . Gains from
                                                                   1      2
trade that are due to improved trading efficiency between markets given these fundamentals
are reflected in the random variable νt . Consequently, we are interested in assessing whether
the average value of νt fell as a result of the market’s expansion.

   Although the autarky prices p∗ and p∗ are unobserved, they characterize willingness-to-
                                1      2
sell in each market: p∗ = Si (Qd ), where Si is market i’s (inverse) supply curve. See Figure
                      i        i
2. To obtain an estimable model, we take logs of (1) to get

                       ln |p1t − p2t | = ln |S1t (Qd ) − S2t (Qd )| + ln νt .
                                                   1t          2t                         (2)

The first terms on the right-hand side serve to strip the variation attributable to changes in
factor prices and consumer demand from the total variation in observed (log) price spreads.
It remains to specify a model for aggregate supply in each market, Sit (Qd ).

   The appeal of this model is that is yields a convenient decomposition and interpretation
of changes in the efficiency of trade when there are confounding shifts in supply and demand

curves. Let α1 be the difference in mean (log) spreads post- versus pre-expansion:

                                  ln νt = α0 + α1 It + ξt .                                (3)

Let ‘%Δ’ denote the change in the (absolute) average price spread post- versus pre-expansion,
expressed as a percentage of the (absolute) average pre-expansion spread. A little algebra
then decomposes the total percent change in spreads into three separate components:

                                      %Δ = r eα1 Ω − 1                                     (4)

                      post                    post
where r = E[exp(ξt )|It    = 1] / E[exp(ξt )|It    = 0] is (loosely speaking) the unexplained
post-pre “variance” ratio, and

                              E |S1t (Qd ) − S2t (Qd )| |It
                                       1t          2t          =1
                         Ω=                                         .
                              E |S1t (Qd ) − S2t (Qd )| |It
                                       1t          2t          =0

We can interpret Ω as the portion of the total percent change in average spreads that is
attributable to changes in the (exogenous) observables, namely, factor prices and consumer
electricity use, each day. We can interpret r exp(α1 ) as the portion that is ‘unexplained’ by
the observables. If the residual log-spread error ξt is normally distributed, then r simplifies
    2        2
to σξ,post /σξ,pre . An estimate of the percent change in average (absolute) spreads due to the
markets’ integration, holding changes in the exogenous observed factors constant, is then
                                     exp(α1 )    2        − 1.                             (5)

Intuitively, the total percent change in average price spreads following integration is com-
posed of a ‘level’ effect, exp(α1 ), and a reduction in the dispersion of price spreads given by
the variance ratio term. This provides a convenient way to decompose the percent changes
in observed price spread in the components of interest.

   Two technical issues merit note. First, an additional advantage of interpreting the
effects of market integration on price spreads via (5) is that it is a variance-stabilizing
transformation over the relevant range of α1 for our application. In estimation we find this
to be quite significant, reducing the standard error associated with the percent change in
(5), relative to the standard error in α1 , by a factor of three or more.

   Second, as written the model in (2) presents a partial identification issue. To see why,
note that we could subtract a constant c from the error term ln(νt ) and multiply each sup-

ply function by ec and leave the left-hand side of (2) unchanged. Thus, a normalization is
required. In estimation we impose (the sample analog to) c = E[ln νt ], which centers the
disturbance term on zero. The unknown value of c can be recovered post-estimation using
additional information contained in the model, viz., that aggregate supply and aggregate
demand must balance. We discuss this step in the following section. Note that this normal-
ization has no effect on the estimate of α1 using the difference in the average fitted residuals
from (2) post and pre, since the unknown value of c is differenced out.

6.2    Specifying Supply Sit (Q)

We now turn to an empirical model for market-level supply. We start with our knowledge of
the underlying production technology: At the level of an individual plant (generating unit),
electricity production is fixed-proportions (Leontief) in two variable factors of production,
fuel and emissions permits. This implies the marginal cost of an individual plant k is

                                MCkt = HRk ( pf + ERk · pe )
                                              t          t
                                          BTU in      $      tons out    $
                                          MWh out   BTU in   BTU in     ton

where HRk is the plant’s heat rate (inverse thermodynamic efficiency), pf is the price of
fuel, ERk is the plant’s emissions rate (NOx and SO2 production), and pe is the price of
emissions permits. The price of fuel and permits varies over time but does not vary (in
our data) across firms or plants in the same region. If, in addition, the emissions rate is
constant across plants, then the market level marginal cost function for production from
plants of fuel type f can be written as

                                   MCf (Q) = g f (Q) (pf + ERf pe )
                                     t                 t        t

where g f (Q) is a monotonically increasing step function. Empirically, the assumption that
all plants using the same fuel have the same emissions rate is implausible (some have
scrubbers, some do not). However, the empirical consequence of this assumption is likely
to be negligible, as a plant’s total permit costs are small relative to its fuel costs, and the
overwhelming determinant of the marginal cost structure is the first two terms, g f (·)pf .

    In all the markets we study, there are two production technologies that set the market’s
prices: gas- and coal-fired generation.9 We also observe an indicator variable for which fuel
     In eastern Pennsylvania, there are a small number of oil-fired plants still in operation that we ignore
here. We believe they are not relevant for explaining prices at PJM West and Allegheny, which are in central
and southwestern Pennsylvania, except under the most extraordinary circumstances.

type is actually on the margin hourly.10 To combine the two, we assume that market-level
supply is proportional to a market level cost function given by

                                     MCt (Q) =         f   It ptotal g f (Q)

where f indexes fuel types c (coal) and g (gas), It = 1 if and only if type f is the marginal
(price setting) technology at t, and the factor price and emissions-related terms are con-
densed to
                                         ptotal = (pf + ERf pe ).
                                          t         t        t

To complete the specification, we need to match the frequency of the price data. These
cover either 16 hour (peak) or 8 hour (off peak) delivery durations. We observe the retail
electricity consumption data at an hourly frequency, and aggregate up to daily (16 or 8
hour) supply explicitly. If we let h index hours and now set t to index the day (price block),
and add a new subscript i that indexes markets, the market-level supply (per megawatt
hour) in market i on day t becomes

                             Sit (Qit ) = γi0 +                    Iih ptotal g f (qih )
                                                                        ih                 (6)
                                                      h∈t f =c,g

where Qit is now a vector of 16 or 8 hourly consumption observations {qih } during day t.

       There are two points regarding implementation that merit note. Although the produc-
tive efficiency function g f is technically a monotone step function, there are hundreds of
individual generating units in our markets. This makes the differences between g f and a
smooth polynomial approximant to it small, especially over the relevant range of quantities
we observe. In estimation, we find no evidence that third or higher-order terms have any
effect on our empirical results; thus, the results reported below are based on second-order
polynomials for g f .

       Second, the plant-level data (discussed in Section 5.2) and regulatory reports from PJM
indicate that the marginal seller’s markup over its marginal cost is typically small in the
markets we examine (a few percent). We have not attempted to estimate markups explicitly.
Instead, any differences between marginal cost and willingness to sell that actually affect
the market price will be absorbed into the estimated productive efficiency function, g f .
Our assumption that this technology-dependent function g f is time invariant becomes not
entirely innocuous as a result; one possible concern is that, coincident with (or as a result
of) of the market’s expansion, suppliers may have changed their behavior and that in turn
       These data are courtesy of the PJM Market Monitoring Unit.

may affect price spreads. The plant-level data help inform this, and we address it further

   Supply Model Data. We briefly summarize a few features of the data that enter the
supply function here. For prices set by the organized market, the variable qih represents
the total retail electricity consumption in hour h of consumers served by local distribution
utilities that are members of PJM (initially). For prices set by the bilateral market, the
same applies for utilities that are not initial members of PJM. These were obtained from
PJM and the major utilities serving load in the Midwestern regions; a complete list and
technical details are given in the Appendix.

   Second, there is the possibility that the marginal fuel indicator Ih might be endogenous
in the price spreads model (2). The reason is that successful arbitrage (between Midwest
producers) and buyers on the East cost may tend to displace gas-fired production and make
coal more likely to be the marginal fuel (in PJM). Although we think the extent to which
this occurs in minor, we nevertheless instrumented for It in the PJM supply specification.
For instruments we projected observed It onto an array of time dummies (by hour of day,
month, season, and their interactions), giving us predicted marginal fuel indicators that are
functions of time alone.

   For the Midwestern bilateral market prices this problem does not arise, as the plant-
level data indicate the marginal fuel is always coal. Further details about fuel data sources
and their construction are provided in the Appendix, including fuel prices, point sources,
delivery costs, and plant emissions rate information.

6.3      Estimated Effects of Expansion on Price Spreads

Table 2 summarizes the estimated model’s results with respect to price spread changes.
The raw parameter estimates for the fitted supply functions in each market are given in
Appendix Table [X]. We discuss the fitted price spread change results here, and the supply
function results in Section 7.

   Column (1) of Table 2 reports the straight pre- versus post-expansion average (absolute)
price spread changes, as a percent of the average (absolute) spread during the pre- period.
These are the same percentage changes reported in our interpretation of Table 1, and are
provided here as a benchmark for subsequent comparisons. The results here and in column
(2) employ the same six months of data pre- and post-expansion as in Table 1. Price
spreads for the AEP-Dayton versus PJM delivery point contrasts are omitted from Table

2, because data non-availability prevents us estimating the results in columns (3) and (4)
(The AEP-Dayton bilateral market prices were not collected systematically prior to 2004).

   To obtain the results in column (2), we fit the supply function model specification (2)
separately for each market-specific pricing point pair shown. This was done by nonlinear
least squares. Using the fitted models, we can now identify the portion of the total percent
change in price spreads (shown in column (1)) that is not attributable to changes in fuel
prices, emissions permit prices, or variation in retail electricity consumption. This ”unex-
plained” change in the price spread after the market’s expansion, evaluated using equation
(5), is shown in column (2).

   The striking feature of these results is the fact that factor price and retail consumption
variation account for very little of the observed changes in price spreads. In the peak-period
prices shown in Panel A, the estimated percent changes in average spreads in columns (1)
and (2) are nearly identical. In the off-peak prices shown in Panel B, the estimated percent
changes in spreads net the effect of observables (column 2) are 14 to 26 percentage points
lower than the total percentage changes in column 1. This indicates that perhaps a quarter
of the total decline in price spreads off-peak may be attributable to changes in weather,
consumption, or factor prices that would have lowered the price spreads in any event.

   To help understand why variation in suppliers’ costs—which rose significantly in late
2004—had little apparent effect on market prices spreads, we examined to the underlying
data. In general, only asymmetric changes in suppliers’ marginal costs should affect price
spreads, even in a commonly congested network. By examining the raw fuel price as well as
weather data for cities near these delivery points, one obvious explanation emerges: While
there are large changes in these variables over time, these changes rarely asymmetric. The
cost of coal delivered to central Pennsylvania producers near the PJM Western Hub pricing
point and the cost of coal delivered to southern Ohio producers near the Cinergy Hub pricing
point tend to move together. This is sensible: Coal is shipped to both sets of generators
(primarily) from the same Appalachian coal mining regions, so the difference in fuel price
levels between the two regions largely reflects transportation costs—which are stable over
this period.

   A similarly set of observations applies to retail electricity consumption, the day-to-
day variation in which is caused almost entirely by weather (Reiss and White, 2007). Daily
weather data indicate that weather in Pittsburg (near the PJM Allegheny delivery area) and
in Columbus (near the AEP-Dayton delivery area) are not appreciably different. Ultimately,
the five delivery point areas in listed in Table 1 are simply not far enough apart to experience

asymmetric changes in retail electricity consumption, or suppliers’ input factor prices.

   In sum, there is scant support for the conjecture that price spreads would have con-
verged if the markets’ expansion had not occurred, due to changes in producers input
factor prices or due to variation in retail electricity demand. Given that the technology of
power production and the determinants of suppliers’ marginal costs are well understood, it
becomes difficult to imagine what other exogenous factor could have altered the markets’
price spreads, coincident with the organized market’s expansion date, by the magnitude
observed in the data.

6.4     Comparison Year Contrasts

To help further rule out possible confounding factors, we have also examined the variation in
price spreads during a comparison year. There were no changes in the market’s organization
(of any kind) during this comparison year.

   In addition, this comparison will net out any possible seasonal effects on price spread
changes. Our pre- versus post-expansion comparisons in Table 1, and columns 1 and 2 of
Table 2, are based on one year of data centered on the PJM expansion date of October
1, 2004. This means that the six months of pre-expansion data span the summer months
(April to September), while the post-expansion data span the winter months (October to
March). As a general matter, there is no obvious reason to suspect a seasonal effect to
price spreads (i.e., why would firms be worse at arbitrage in the summer than in winter?).
Nevertheless, we addressed this possibility using a comparison year contrast.

   To match the seasonal change in columns 1 and 2 exactly, we selected a comparison
period of 12 months centered on October 1, 2003. We then re-estimated the model using
data for the entire two year span from April 1, 2003 to March 21, 2005. As noted above, the
price data for the AEP-Dayton bilateral market delivery point do not extend back beyond
2004, so are not included in this analysis. The other two bilateral market delivery points
are not available much earlier than 2003, which determined our choice of the comparison
year period.

   For the results reported in column (3) of Table 2, we estimated a benchmark comparison-
year contrast specification without a control function for factor prices and retail consump-
                                             Post       Year2       Winter
                  ln |p1t − p2t | = α0 + α1 It    + α2 It     + α3 It      + εt

                     Post = 1 for t on or after October 1, 2004 (post-expansion), I Year2 = 1
where the indicator It                                                             t
                                      Winter = 1 if t falls in the months of October through
for t on or after April 1, 2004, and It
March (inclusive) in either year. The coefficient α1 therefore measures the change in (log)
spreads between the six months pre- versus post-expansion, relative to the change in log
spreads between the corresponding six-month calendar periods the prior year. We then
transformed the estimated value of α1 into an estimated percent change (here relative to the
prior year’s change), using (5). This was done separately for all delivery point contrasting
pairs listed in Table 2, peak and off-peak.

   The estimated percent change in spreads, relative to the comparison year change, is
shown for each contrasting delivery point pair in column (3). These results are little changed
from those in columns (1) and (2). The reason is straightforward: The comparison year
data show little difference in the average (absolute) price spread over the six months after
October 1, 2003, relative to the six months before October 1, 2003. We infer that without
a structural change in the markets’ organization, changes in demand between summer and
winter do not result in changes in between-market price spreads.

   Things are slightly different in column (4), but the same conclusion emerges. Here we
first estimated the model in equation (2) by nonlinear least squares, using the full two years
of data. We then obtained comparison-year contrasts for the unexplained (log) spread by
linearly projecting the residuals from the fitted model (2) onto the time-period indicators
used for column (3), or

                                       Post       Year 2       Winter
                      ln νt = α0 + α1 It    + α2 It      + α3 It      + ξt                 (7)

This was again done separately for each contrasting delivery point pair listed in Table 2, peak
and off peak. We then used (5) to obtain point estimates of the percent change in spreads
relative to the change during the comparison year, net the effects of any simultaneous
changes in factor prices or retail energy consumption levels.

   As with the straight pre- versus post-expansion comparisons, controlling for variation
in costs and retail consumption results in modest changes in estimated effect of integration
on price spreads. Interestingly, the estimated percent changes in spreads (net the effects of
the observables) are slightly larger than in columns (1)-(3) for the peak periods in Panel
A, and slightly smaller than in the other columns for the off-peak periods in Panel B. In
addition, the explanatory power of the control function increases dramatically; while the
R2 values for models in column (3) are typically about .1, the R2 values for the models
in column 4 all range from .6 to .7. Thus, the control functions are helping to explain a

portion of the observed spread changes, when contrasted against the relationship between
these factors and price spreads during the comparison year.

     Summary. In sum, by combining the initial price spread evidence from Section 4 with
the analyses performed here and summarized in Table 2, six major observations emerge:

    (i) Price spreads between markets fell dramatically after the organized market’s expan-
       sion, for each of the four contrasting delivery points, in both peak and off-peak periods;

 (ii) The magnitude of price spread convergence is substantial, both relative to pre-expansion
       price spreads and relative to the comparison-year spread changes;

(iii) These changes are apparent within days of the expansion, and price spreads remained
       smaller persistently thereafter;

(iv) This convergence of price spreads cannot be explained by changes in factor prices, or
       by demand conditions;

 (v) Nor is there any evidence that price spreads change systematically at the same calen-
       dar times in prior years, when no structural changes in market organization occurred;

(vi) Nor were there any systematic changes in transmission or productive capacity during
       the time span studied here.

     On this basis, we led to conclude that the expansion of the organized market caused the
substantial change in wholesale electricity market prices we observe in these data. In addi-
tion, the quantity evidence in Section 5 suggests that these price changes were accompanied
by real changes in which firms and production facilities generated the power consumed by
many (indeed, millions) of consumers. Taken together, these results imply that the or-
ganized market design was able to identify gains from trade—reallocating production to
the lowest-cost producers—that were not being realized by the previous bilateral market

7      Welfare

The model presented in Section 2 enables us to estimate how large are the newly-realized
gains from trade after the organized market’s expansion. The central task the model enables
us to address is how trade differed from what would have happened if, counter to fact, the
market’s expansion had never occurred.

   To give our method for calculating efficiency gains a clear conceptual footing, it is useful
to return to the simple two-sector model of trade from Section 6. Let Δqt be the actual

quantity traded between two market-specific delivery points on day t, for t after October 1,
2004. Similarly, let Δqt be the counterfactual quantity that would have been traded after

October 2004 ‘but for’ integration. Because price spreads converged post-integration, we
know that |Δqt | > |Δqt |, that is, traded volume increased as a result. Our analysis is based
             a        c

on how suppliers’ willingness to sell at different market-specific delivery points varies over
the range of quantities between Δqt and Δqt .
                                  a       c

   In any time period t after October 1, 2004, the welfare gain attributable to integration
is the difference in sellers’ valuations for the incremental quantities traded:
                             Δqt                             a
                    Wt =           S1t (Qd
                                         1t   − θ) dθ −          S2t (Qd + θ) dθ
                                                                       2t                 (8)
                            Δqt                             c

The switch of the limits of integration for each market, along with the sign on θ, reflect only
sign conventions regarding the direction of trade (Δqt and Δqt are positive for net flows
                                                     a       c

into market 1). In the context of Figure 2, the value Wt is the difference between the shaded
area depicted in the figure and the larger shaded area that would apply if, counterfactually,
the price spread was at the higher level that would have prevailed if the market’s expansion
had not occurred. We will estimate Wt separately for each day and each price block (peak
and off peak), then sum these values to obtain total welfare over T -period horizons.

   As written, (8) gives us a measure of ‘pairwise’ welfare gains between the markets, that
is, for a specific pair of delivery points. It will differ (slightly) between the pairs, because
supply functions are not identical everywhere. Because of this, any pairwise value of Wt
will be less than the value we would obtain if we used four supply functions at the same
time, which would be a better measure of total welfare. (We will address this in the next

   Implementation. To evaluate the gains from trade using (8), we need to estimate
the counterfactual price spreads |Δpc | and net flows Δqt that would have applied for t

after October 1, 2004, ‘but for’ integration. In addition, we require estimates of the actual
between-market flows for the delivery point pairs where these data are not directly observed.

   We proceed in two steps. First, we use the fitted spreads model to predict the coun-
terfactual price spread each day t. Second, we then solve the model ‘in reverse’ for the
quantity traded (Δqt ) that yields the counterfactual price spread. We replicate this second
step with the actual price spreads after the market’s expansion, providing the estimate of

Δqt . Because our supply function specification (6) is a polynomial in quantity, the integral

in (8) is simple to evaluate once we determine the limits Δqt and Δqt .
                                                            a       c

   The first of these steps is straightforward. To estimate the counterfactual price spread,
we use the comparison-year contrast model based on (2) and (7). If the market’s expansion
had not occurred, then the “unexplained” drop in price spreads relative to the comparison
year change, or α1 in (7), would be zero. Setting α1 = 0 implies a counterfactual price
spread of

                                                            Year 2       Winter
            |Δpc | = |S1t (Qd ) − S2t (Qd )| · exp α0 + α2 It
               t            1t          2t                         + α3 It      + ξt     (9)

which is equivalent to
                                     Δpc = Δpt / exp(α1 ).
                                       t                                                (10)

where Δpt is the observed price spread. The absolute values are unnecessary in (10) as the
price difference on each side always has the same sign. We obtain the estimated counterfac-
tual price spread by plugging in the estimate of α1 from the fitted models in Section 6. The
counterfactual price spreads therefore vary (as indicated in Table 2) by delivery point pair
and period (peak and off peak). Note that, conceptually, here we are (implicitly) assuming
that the same disturbance term ξt that actually occurred on date t post-expansion would
have also applied on that date had the expansion not occurred. This seems sensible, as the
main random factors that we cannot account for in our model that might affect spreads
(network line failures, generator forced outages large enough to move prices, and the like)
should not be assumed away in the counterfactual case of no market expansion.

   To obtain the quantities traded, we use a second implication of the two-sector imperfect
trade model from Section 6. Specifically, the price in each market is given by willingness to
sell (inverse supply) evaluated at the autarky quantity, less net imports or exports:

                                     p1t = S1t (Qd − Δqt )
                                                 1t                                     (11)

                                     p2t = S2t (Qd + Δqt )
                                                 2t                                     (12)

Here Δqt > 0 for exports from market 2 to 1, and negative for imports from 2 to 1.
Subtracting gives
                           Δpt = S1t (Qd − Δqt ) − S2t (Qd + Δqt )
                                       1t                2t                             (13)

This is a ‘markets-clear-in-quantities’ condition. It must hold for any set of price spreads;
all it says is that imports into market 1 from market 2 must equal exports from market 2
to market 1. Since that must hold for any price spread, it also defines the counterfactual

quantity, Δqt , that would yield the counterfactual price spread, Δpc .

   In principle, solving (13) for Δqt using the observed price spreads yields the quantity
  a   needed to evaluate welfare using (8). Repeating the process using the counterfactual
price spreads yields the counterfactual traded quantity Δqt that we predict would have

transacted had the market’s expansion not occurred.

   In practice, there is a small wrinkle. Recall that the model in (2) identifies each market’s
supply function only up to scale. The normalization imposed during estimation implies that
the actual and estimated supply function are off by an unknown constant ec , or

                                  Sit (Q) = ec Sit (Q) + error

Estimates of welfare using quantities obtained from (13) using Sit (Q) would therefore tend
to be systematically off by a factor of ec .

   To handle this, we can use the ‘markets-clear-in-quantities’ condition to estimate c along
with Δqt . Now (11)-(12) become

                               p1t = λ S1t (Qd − Δqt ) + error
                                             1t                                          (14)

                               p2t = λ S2t (Qd + Δqt ) + error
                                             2t                                          (15)

where λ = exp(−c). Since p1 and p2 are observed, we know everything except for the
value of λ and the T values of Δqt . That means (14) and (15) comprise 2T equations in
T + 1 unknowns. Solving jointly via least squares provides c as well as estimates of the
actual quantities for each day, Δˆt , that were not observed in our data. Last, we obtain
the counterfactual quantities by solving (13) for Δqt using the counterfactual prices on the
left-hand side and the corrected supply function estimates λSit for Sit on the right.

   Results. We use the fitted models to evaluate the gains from increased trade between
each pair of delivery regions listed in Table 2, for both peak and off peak periods. Aggregated
to an annual basis, the efficiency gains we estimate range from $66 million to $97 million
across the four contrasting delivery points. These should not be summed together; rather,
because we have evaluated the gains from trade pairwise (as opposed to solving for the
implied flows between all delivery points simultaneously), these should be interpreted as
providing four different estimates of the total gains from improved trade between all of
these delivery regions.

   In general, the counterfactual price spreads we obtain from (10) are quite similar to the
price spreads that we observed prior to the market’s expansion. This is expected, given the
results in Table 2; there, we found that little of total change in price spreads pre- versus
post-expansion could be explained (statistically) by the increase in producer’s factor prices
that occurred in the winter of 2004.

   The estimated changes in quantities traded between regions (that is, Δqt − Δqt ) are all
                                                                          a     c

of highly plausible magnitudes. For peak periods, we estimate that the organized market’s
expansion increased average hourly flows between the Michigan-First Energy delivery area
and each of the PJM delivery areas (Western Hub and Allegheny) by 730 and 840 megawatts;
on-peak flows from the Cinergy area increased by 1600 megawatts to Allegheny and 1200 to
Western Hub, on an average hourly basis. Overall, the elasticity of net flows is very similar
between all delivery points; for every additional 390-450 megawatts transferred between
delivery points, the price spread between them falls by $1 per megawatt. The larger flows
we estimate from Cinergy reflect the fact that the decline in price spreads between the PJM
delivery points and Cinergy was roughly double the decline with Michigan-First Energy.
These price spread changes mirror their changes in the pre- versus post-expansion price
spreads shown in the descriptive data; see again column (3) of Table 1, Panel A.

   For the off-peak periods, the counterfactual price spread changes are significantly larger.
Again this is similar to the pre- and post-expansion spread convergence summarized in Ta-
ble 1. Consequently, the changes in quantities shipped between regions are nearly double
those on peak. We estimate the organized market’s expansion increased hourly average
flows during off-peak periods by between 1200 and 2300 megawatts, depending on the de-
livery point pair. The largest quantity changes are observed between Cinergy and the PJM
Allegheny regions, which have the largest transmission network interconnection capacity,
and which exhibited the largest convergence in off-peak price spreads (see again Table 1,
Panel B).

   Cost of Market Expansion. By any measure, these are large efficiency gains following
the adoption of the organized market’s design. As noted in the introduction, however,
there are costs to implementing a new system of market organization. These costs can be
compared to the efficiency gains reported above, providing a better assessment of the net
benefits of expanding the organized market design.

   The costs of implementing the new market design were incurred by two sets of market
participants: The market operator itself (PJM), and the individual firms that joined the
market. Regulatory accounting filings prepared by PJM for its members and the FERC

report total expansion expenses of $18 million, through 2005. These are one-time, non-
recurring expenses due to the expansion of the market. For the new members, accounting
data filed with the SEC by American Electric Power indicate internals costs of re-organizing
its wholesale market operations due the PJM expansion of $17 million; forward-looking
statements characterize this as a one-time expense. Other market participants’ expenses
are more difficult to obtain, but based on volume-of-production and trading data, and the
fact that all other new members relied upon AEP’s regional transmission network prior to
the market’s expansion, we believe are likely on the order of $4-5 million. In total, this
amounts to approximately $40 million in one-time implementation costs of expanding the
market’s design.

    Combining these benefits and costs, the picture that emerges is that for an initial in-
vestment of approximately $40 million the participants in these markets realized welfare
gains of $66 to $97 million over the first year alone. At the usual risk of extrapolation, if
gains of this magnitude in subsequent years are of similar magnitude, the present value to
society of expanding this organized market’s design is remarkably large.

8    Conclusion

Our motivation for this paper arose from a vigorous—and, we believe, poorly informed—
policy debate about the merits of organized market designs in liberalized electricity markets.
This debate reflects two distinct, but related difficulties that confront policy makers. First,
the potential for a more efficient market design to substantially reallocate production from
high-cost firms to lower-cost competitors provides a political incentive for the market par-
ticipants who stand to lose to oppose it. Second, there is the technocratic challenge that
the theoretical appeal of a well-specified market design must be balanced against the cost of
implementing it. Given these incentives and challenges, it is not surprising that a consensus
among industry participants and policy makers on this fundamental trade-off has proved

    The central contribution of this paper is to provide a detailed empirical assessment
of this question. The expansion of the organized market design used by the PJM Inter-
connection into the Midwest in 2004 provides one of the only opportunities to address
this quantitatively. As industry participants within this region are well aware, there were
dramatic changes in market outcomes after the expansion: lower-cost facilities increased
production, price spreads between Midwestern and Eastern delivery points converged, and
the quantities of power transferred between them increased substantially. These findings are

consistent with the theoretical economic concern that decentralized bilateral markets may
have difficulty achieving efficient allocations of the numerous complementary services—viz.,
generation and transmission—required to trade electrical power with utmost efficiency.

   One perspective that bears comment here relates to the behavior of the firms new to
the organized market system. Specifically, perhaps the efficiency improvements we have
pointed to here arose because the expansion of the organized market led the new market
participants to change their willingness to supply. We alluded to this possibility when
discussing our interpretation of the supply specification model in section 6.2. Stated in
other words, perhaps the firms that joined PJM simply decided to offer their production at
lower prices (that is, by bidding more aggressively) into the organized market, relative to
their previous supply behavior bilateral market.

   We are skeptical of this possibility, for several reasons. First, from a theoretical per-
spective, it is difficult to conceive why such a change in willingness-to-sell would be profit-
maximizing behavior. The identities and number of firms operating in these markets was
the same throughout the period we study, and—if the bilateral markets were not subject
to trading imperfections—then the new exchange members would have faced the same set
of trading opportunities before and after the organized market’s expansion. Second, there
is the empirical fact that prices in the delivery region where the new members physical pro-
duction assets are located increased sharply following the market’s expansion. This fact is
inconsistent with a reduction in willingness to sell by producers in this region, but consistent
with an increase in demand for the power they produce (from buyers to the east).

   Third, our results indicate that the quantities delivered to the two main PJM delivery
regions from producers to their west nearly tripled post-expansion, but these two PJM deliv-
ery regions’ price levels fell only about ten percent. This is an extraordinarily large elasticity
response, although perhaps that is to be expected in an homogeneous-good market. Empir-
ically, it would not have been profitable for the new exchange members to produce as little
as they actually did before the market’s expansion—unless bilateral market imperfections
obscured the trading possibilities subsequently identified by the organized market’s design.

   We are led to the seemingly inexorable conclusion the expansion of the organized market
design identified new trading opportunities that were simply not realized by the bilateral
trading system that preceded it. The magnitude of the market changes analyzed here clearly
call into question the assertion that organized market designs are not worth their costs of
implementation. It is worth noting that in 2005, the year after the organized market’s
expansion studied here, several other large utilities that previously traded only in bilateral

market venues joined the PJM market system. Thus, while the evidence we have brought
together here may prove useful to public policy debates over the merits of organized market
designs, it seems clear that many participants in the industry—who directly witnessed the
changes after the 2004 market expansion documented here—have already concluded this
evidence is sufficiently compelling to act upon it.

Appendix A.



Borenstein, Severin, James B. Bushnell, and Frank A. Wolak (2002). “Measuring Market Ineffi-
  ciencies in California’s Restructured Wholesale Electricity Market.” American Economic Review
  92(5): 1376-1405.
Bushnell, James B., Erin T. Mansur, and Celeste Saravia (2006). “Vertical Arrangements, Market
  Structure, and Competition: An Analysis of Restructured U.S. Electricity Markets.” University
  of California Energy Institute, CSEM Working Paper 126.
Joskow, Paul L. (2001). “California’s Electricity Crisis.” Oxford Review of Economic Policy 17(3):
Hurwicz, Leonid (1973). “The Design of Mechanisms for Resource Allocation.” American Economic
  Review 63(2): 1–30.
Klemperer, Paul (2002). “How Not to Run Auctions: The European 3G Telecom Auctions.” Euro-
   pean Economic Review, European Economic Review 46(4-5): 829-45.
Olson, Mark, Stephen Rassenti, Mary Rigdon, and Vernon Smith (2003). “Market Design and
   Human Trading Behavior in Electricity Markets.” IEE Transactions 35: 833–49.
Mansur, Erin T. (2007a). “Upstream Competition and Vertical Integration in Electricity Markets.”
  Journal of Law and Economics 50(1).
  Mansur, Erin T. (2007b). “Measuring Welfare in Restructured Electricity Markets.” Forthcoming
  in Review of Economics and Statistics.
McMillan, John (1994). “Selling Spectrum Rights.” Journal of Economic Perspectives 8(3): 145–162.
Roth, Alvin E. (2002). “The Economist as Engineer: Game Theory, Experimentation, and Compu-
  tation as Tools for Design Economics.” Econometrica 70(4): 1341–1378.
Rothschild, Michael (1973) “Models of Market Organization with Imperfect Information: A Survey.”
  Journal of Political Economy 81(6): 1283–1308.
Smith, Vernon L. (1962). “An Experimental Study of Competitive Market Behavior.” Journal of
  Political Economy 70(2): 111–137.
Smith, Vernon L. (1964). “Effect of Market Organization on Competitive Equilibrium.” Quarterly
  Journal of Economics 78(94): 181–201.
Wilson, Robert A. (2002). “Architecture of Power Markets.” Econometrica 70(4): 1299–1340.
                                                     TABLE 1
                                                                          Average Prices for Day-Ahead Forwards ($ per MWh)
               Contract Delivery Point                             Pre-Expansion      Post-Expansion               Post – Pre      Std. Error of
               (and approximate location)                        (Apr. - Sep. 2004) (Oct. '04 - Mar. '05)          Difference      Differencec

                                                             Panel A:    Peak-Period Delivery b
               Exchange-based Prices
                    PJM Western Hub (Pa.)                               50.98                   50.71                 -0.27         (1.79)
                    PJM Allegheny (Pa. and W. Va.)                      50.41                   49.80                 -0.60         (1.92)
               Bilateral Market Prices
                      MichFE (N. Ohio valley)                           46.00                   47.78                 1.78          (1.77)
                      AEP-Dayton (C. Ohio Valley)                       43.41                   45.81                 2.40          (1.87)
                      Cinergy (S. Ohio Valley)                          43.59                   46.33                 2.75          (1.80)
               Price Spreads Between Markets (average absolute price differences)
                      PJM Western Hub v. MichFE                          6.23                    4.31                 -1.93         (0.72) ***
                                      v. AEP-Dayton                      8.28                    5.23                 -3.05         (0.96) ***
                                      v. Cinergy                         8.16                    5.18                 -2.99         (0.86) ***
                      PJM Allegheny v. MichFE                            5.75                    4.22                 -1.53         (0.76) **
                                    v. AEP-Dayton                        7.68                    4.89                 -2.79         (1.00) ***
                                    v. Cinergy                           7.56                    5.02                 -2.54         (0.90) ***

                                                          Panel B:      Off-Peak Period Delivery b
               Exchange-based Prices
                    PJM Western Hub (Pa.)                               27.88                   31.98                 4.10          (1.54) ***
                    PJM Allegheny (Pa. and W. Va.)                      28.01                   30.61                 2.60          (1.35) *
               Bilateral Market Prices
                      MichFE (N. Ohio Valley)                           18.32                   29.15                10.83          (1.10) ***
                      AEP-Dayton (C. Ohio Valley)                       17.32                   27.98                10.66          (1.20) ***
                      Cinergy (S. Ohio Valley)                          16.99                   28.41                11.42          (1.17) ***
               Price Spreads Between Markets (average absolute price differences)
                      PJM Western Hub v. MichFE                          9.58                    3.80                 -5.78         (0.70) ***
                                      v. AEP-Dayton                     10.60                    4.54                 -6.06         (0.82) ***
                                      v. Cinergy                        10.90                    4.31                 -6.58         (0.76) ***
                      PJM Allegheny v. MichFE                            9.70                    2.80                 -6.90         (0.59) ***
                                    v. AEP-Dayton                       10.72                    3.40                 -7.32         (0.67) ***
                                    v. Cinergy                          11.02                    3.29                 -7.73         (0.62) ***

               Notes. (a) Delivery points for electricity transactions are defined by area of the high-voltage transmission grid, not single points
               on a map. The locations above correspond to contiguous geographic regions, as follows (approximately): PJM Western Hub is
               central and western Pa.; PJM Allegheny is southwestern Pa. and northern W. Virgina; AEP-Dayton is central Ohio and southern
               W. Virginia; MichFE is northern Ohio and lower Michigan; and Cinergy is southern Indiana and southwestern Ohio. (b)
               Separate contracts are traded for peak (7am to 11pm) and off-peak (11pm to 7am) delivery. (c) Newey-West standard errors
               assuming a five-day lag structure. Significance indicated for 1% (***), 5% (**), and 10% (*) levels.

                                                                           Figure 1.
                                                             Energy Transfers From AEP Region to
                                                PJM West / PJM Allegheny Regions, As a Share of Transfer Capacity

Transfer Capacity Utilization Rate





                                       Oct-03   Jan-04       Apr-04      Jul-04      Oct-04      Dec-04       Apr-05
                                                     TABLE 2
                                       Percent change in average daily absolute price spreads between markets,
                                  for eight delivery point pairs by delivery period. Standard errors in parentheses. a

                                                                                Pre v. Post                 Difference-in-Difference
                    Exchange v. Bilateral Market Contrast                 (1)             (2)                   (3)              (4)

                    Controls for Factor Price                             NO             YES                   NO               YES
                       and Demand Variation?

                                                            Panel A:     Peak-Period Delivery c

                    1.   PJM Western Hub − MichFE                       -0.34             -0.28                 -0.36            -0.52
                                                                       (0.11) ***        (0.13) **             (0.17) **        (0.13) ***
                    2.   PJM Western Hub − Cinergy                      -0.41             -0.35                 -0.48            -0.68
                                                                       (0.10) ***        (0.11) ***            (0.13) ***       (0.08) ***
                    3.   PJM Allegheny − MichFE                         -0.37             -0.34                 -0.43            -0.62
                                                                       (0.11) ***        (0.13) **             (0.16) ***       (0.11) ***
                    4.   PJM Allegheny − Cinergy                        -0.39             -0.41                 -0.48            -0.65
                                                                       (0.11) ***        (0.10) ***            (0.13) ***       (0.09) ***

                                                          Panel B:     Off-Peak Period Delivery c

                    5.   PJM Western Hub - MichFE                       -0.72             -0.58                 -0.67            -0.60
                                                                       (0.04) ***        (0.09) ***            (0.09) ***       (0.12) ***
                    6.   PJM Western Hub - Cinergy                      -0.74             -0.48                 -0.63            -0.51
                                                                       (0.04) ***        (0.11) ***            (0.09) ***       (0.13) ***
                    7.   PJM Allegheny - MichFE                         -0.80             -0.69                 -0.78            -0.75
                                                                       (0.02) ***        (0.06) ***            (0.06) ***       (0.07) ***
                    8.   PJM Allegheny - Cinergy                        -0.80             -0.61                 -0.72            -0.62
                                                                       (0.03) ***        (0.08) ***            (0.07) ***       (0.10) ***

                    N. Observations                                      261              261                   522               522

                    Notes. Summary results from separate regression estimates for each of eight delivery point price-spread pairs shown.
                    Complete parameter estimates reported in Appendix Table A[XX]. Point estimates shown are (approximate) post-
                    expansion percent changes in average absolute price spreads for day-ahead forward contracts, by delivery period, with
                    indicated controls/contrasts (see text). (a) Newey-West standard errors, assuming a five-day lag structure; significance
                    indicated for 1% (***), 5% (**), and 10% (*) levels. (b) See Table 1 notes for delivery point geographic areas. (c)
                    Separate contracts are traded for peak (7am to 11pm) and off-peak (11pm to 7am) delivery.

                             Figure 2.

$                                                          $




               Q1S                       Q2S

                Demand Q1D                 Demand Q2D

To top