STRATEGIC BEHAVIOUR OF WINNING BIDS IN THE
BRAZILIAN TRANSMISSION AUCTIONS
Amanda Pimenta Carlos*
Joisa Dutra Saraiva+
*Graduate Student at Getulio Vargas Foundation –Graduate School of Economics.
+
Professor, Getulio Vargas Foundation.
Corresponding author: Amanda Pimenta - Email: [email protected]
November 18, 2009
Abstract
Using a sample of transmission lines auctions data, from 2000 to 2008, we looked for an explanation, beyond
the traditional questions about strategic behavior and optimality, to the great di¤erence among the price cap
established for revenue in these auctions and the winning bids. We started with the hypothesis that these
high markdown observed are a consequence of a synergy, given by scale gains in construct and administrate
transmission lines close one to another. Firstly, the analysis was made with reduced-based models, as the
transmission auctions are hybrids, with a First Price Sealed Bid auction at the …rst phase and an Ascending
Open Bid (English) auction at the second phase. The empirical results showed that a plenty of factors have
impact on transmission lines winning bids, as such the number of competitors, technology features of the lines,
the price cap, the necessary investment level for the transmission line, and, mainly, the location of the lines,
indicating that synergies are in fact a relevant constituent of the current Brazilian transmission lines competitive
scenario. We also applied Principal Component Analysis to deal with a possible multicollinearity problem and
estimate a non-linear regression to test a theshrold in the bids. Finally, we model an economic experiment testing
that an alternative design for the Brazilian transmission auctions, with neighbor lines auctioned simultaneously
would improve the rent extraction from the transmission companies.
JEL Codes: D44, Q40, H57.
Keywords: Auctions, Procurement, Econometrics of Acutions, Electric Energy, Strategic Behaviour Analysis
in Incomplete Information Games.
1
1
Introduction
Stability and con…ability of electric energy sector is a determinant question for society and, because of this, it is
to Brazilian academy too. The importance of this sector to a sustainable growth is undeniable and the recent
experience of the Power Rationing Crisis in 2001 showed so troublesome can be the planning and execution of
policies in the refered scope of electric energy.
The electric energy transmission lines 1 , specially, are indispensable to conect distant points to the generation
plants - it is a cheap and quick way to attend electricity demand in these places, compared to instalation of new
generation plants - and to give support to the demand in the largest consumers locations. To have a wide system
of transmission lines is "synonym" of a bigger security in the sector.
Whitin this context, and not the least important to Brazilian electricity sector stability, we can insert the
transmission lines auctions, that are the current way to conced the electric energy transmission construction and
service. From 1998 to 2008, the National Electricity Regulatory Agency (ANEEL) auctioned 81 transmission lines
bundles, that adds 28,263.1 kilometers to the system; 23,132.1 kilometers of this total are already operating and
the remaining is in construction.
In these auctions, a more real tendency, from 2003, is that discounts in the bids in auctions for the transmission
line concession are more and more signi…cant. An example of this extension is that the last transmission line auction,
accomplished in December 2006, involved the o¤ering of six lots, organized by ten lines and three substations, with
1,033 kilometers in total. The average discount was 49.36%. Yet in the auction accomplished in November which
involved 14 lines and three substations, in a total of, approximately, 2.25 thousand kilometers of extension, the
average discount was of 51.1%.
Figure 1 - Di¤erences Between Winning Bids and Reserve Prices
Brandão & De Castro (2006) built a model to try explaining the cause of these high discounts. According to the
authors, the higher discounts are intrinsically related to the drop of the "Brazil Risk", that is, with the improvement
in the perception of the national investments to the foreigners. They analyze:
"The improvement in the indicators of the country’s external solvency, helped by international liq1 It’s worth mentioning that Brazilian electric energy sector activities are grouped in four principal segments:
(i) Generation;
(ii) Transmission;
(iii) Distribution;
(iv) Comercialization.
2
uidity has a strict link with the cost and viability of the investments in infrastructure in Brazil", they
stated in the article. "In the moment of the investment evaluation, the risk perception of a country
is included in the cost of the capital, as much in the calculations of the Brazilian companies as in the
foreigners".
However, both authors understand that only the drop of the Brazil Risk does not explain all the "aggressiveness"
of the transmission auction proposals. Part of this change in the tendency is also a result, according to them, of a
higher optimism, especially in relation to the perspectives of future …nancing via BNDES and capital market.
On the other hand, ANEEL states that discounts have demonstrated that the adopted systematics is well
succeded, in such a way that competition in the transmission process results in cheaper taxes to the consumer.
In spite of the levels of high discounts observed, the market does not consider that investors will abandon the
transmission sector. On the contrary, they have been more and more numerous.2 This apparent contradiction smaller return taxes and longer terms that attract investors - is explained by the sector structure, which presents
clear rules, for example. Regulatory mark stability has been an expressive point in the attraction of investments in
this area.
As presented above, the auction is won by the proponent who o¤ers the higher discount to the Allowed Annual
Income3 . Discounts have been higher and higher in the auctions accomplished till the present date and analysts
cannot explain the extent reached for this di¤erence between the estimated value by ANEEL as ideal to the construction, maintenance and operation of the transmission lines and the o¤ered value by the transmission companies
in the auctions.
According to Brandão & De Castro (2006), the main indicator to evaluate the price in the auctions is the ratio
between the winning Allowed Annual Income (RAP) and the investments necessary to build the line (demonstrated
in the investment budget provided by the winner of the auction). In the …rst auctions, the RAP/Investments was
about 20%. However in the auction of November 2005, a lot reached an indicator equal to 9.6%.
The authors point out the following factors as an explanation to the continuous drop in the return required by
the transmission companies:
(i) Low risk;
(ii) Part of the explanation is related to the regulatory mark stability of the electric sector, especially in the
transmission segment;
(iii) Another factor is the reduction of the third party capital, evident by the drop trajectory of the Long Term
Interest Rate (TJLP);
(iv) The main reason of the drop in the remuneration demanded by the energy transmission is the cost reduction
of the own capital (because it is common to involve the variable for country risk in this analysis).
An article from Francellino & Polito (2007) indicates that Spanish transmission companies are the responsible
companies for the higher discounts in the transmission auctions. The reason of such aggressive strategy by the
Spanish companies is not a matter of simple answer, but otherwise, it is a set of advantages provided by the
business structure of these companies, from facilities to get credit to tributary incentives.
2 According to Ribeiro (2007), there are around 50 transmission companies in Brazil, and 30 of which, approximately, entered in the
segment after the sector restructuring.
3 The Allowed Annual Income, just to stand out, is obtained by 15 years. After this period, its value falls to the half. The readjustment
to this income is annual, based in the IPCA, being the single discount for line unavailability.
3
In spite of the fact we understand the in‡uence of the emphasized arguments, the aim of this paper is to search
for justi…cations for the behaviors of the observed bids in the transmission auctions from the Auction Theory, in
its empirical line. Thus, we will analyze, based in the empirical models of auctions and biddings in use nowadays,
the factors that contribute to the distance of these bids in relation to the reserve price - or Allowed Annual Income
- …xed by the regulatory agency. We assume the hypothesis in this paper that there is scale gains by managing
and operating transmission lines next one to another. Auction winners’strategic behavior analysis will allow us to
point out what are the other features of these auctions that a¤ect the victorious bid.
In our empirical exercises we found evidences that many factors impact the winning bid in the transmission lines
auctions, as the number of competitors, the extension of the lines lines, the reserve price, the necessary investment
level for the transmission line, and, mainly, the lines location, indicating that synergies are in fact a relevant
constituent of the current Brazilian transmission lines competitive scenario. We also applied Principal Component
Analysis to deal with a possible multicollinearity problem and non-linear regression to test a theshrold in the bids.
Finally, we model an economic experiment testing that an alternative design for Brazilian transmission auctions,
with neighbor lines auctioned simultaneously would improve the rent extraction from the transmission companies.
2
The Electric Energy Transmission Lines in Brazil
After the restructuring of the electric sector, the private sector started having an important role in the construction,
maintenance and operation of the electric energy transmission lines in Brazil. Such service was made by state
companies previously. According to Hirota (2006), the new model encourages the private participation in a sector
whose initial investments are very high and not recoverable in the short term.
Transmission lines system as a whole is known as National Interconnected System (SIN) and comprises two main
regions: South-Southeast-Center West subsystem and North-Northeast subsystem. Most of the isolated systems,
i.e., not connected to the SIN, belong to the North Region.
According to ANEEL, with transmission line ampli…cation it is possible to o¤er a bigger security in the SIN
operation. Each new line constructed in the country contributes to amplify the electric energy interchange capacity
between regions. Until 2005, almost 83 thousand line kilometers were in operation in the SIN. The construction
of these lines generated nearly 20 thousand jobs and attracted national and international investors, mainly from
Spain, Italy, Argentina, Swiss and the United States.
System National Operator (ONS) is in charge of the system operation and administration, and it is regulated
and inspected by ANEEL. Besides the administration of the line operation, ONS’s role is to propose the necessary
ampli…cation and reinforcement to the Basic Net, based on the demand growing projection and systemic risk.
Hirota (2006) points out that ONS’s coordination in the transmission is essential to optimize SIN, reducing
losses and risks in the present and future. There is energy loss in all transmission and, because of that, the more
is the charge in the system, the more are the losses. This is the reason for the system to be always using its total
capacity, in order to serve the demand in the peak hours.
Sector analysts (in Brasil Energia, 2007) state, for example, that in the crisis supply of 2001, a larger interlink
with transmission lines could have avoided the consumers’rationing and the resultant economic prejudice of the society4 . However, nowadays specialists’opinion is that, with the actual system con…guration - much more interlinked
4 Di¤erently
from today’s situation, in 2001 there was a long term potential gain if the previously described subsystems were inter-
4
than in 2001 - the transmission in any hypothesis could replace the generation.
3
The International Experience in Transmission Lines Investments
According to França and Ramos (1998), after the reestructuring process of deregulation occured in the late 1990s
in the electric system around the world, the transmission sector started to pursue new objectives, as such:
(i) neutrality;
(ii) guarantee of free access;
(iii) guarantee of quality and con…ability to the market;
(iv) guarantee the integration of all the market agents;
(v) allow the energetic optimization;
(vi) minimize the transport tari¤, to facilite business between generators and consumers.
As one can see, these are complex issues and even in the international experience we don’t have yet a satisfatory
solution.
In the Independent System Operator (ISO) of California, United States, for example, they have an indicative
planning for the construction of new transmission lines. However, the decision to expand the system is completely
in the agents hands. The system operator has the role to administrate the transmission constraints.
In England, the expansion planning is made by the National Grid, which has the function of system operator and
electric energy transmission operator. The proposed reinforcement in the transmission are justi…ed to the agents,
consolidated in a report directed to the regulatory agency.
Brazil, for its geographic and regulatory features, had in the recent years the need of enormous transmission
investments. The country had an indicative planning to expand the generation and the transmission, but also have
a determining planning to transmission, elaborated by the national system operator, the ONS, to the horizon of
…ve years. Proposed reinforcements are directed to the regulatory agency (ANEEL). Finally, the ANEEL conduct
the auction process to de…ne the transmission concessionaire that will develop the transmission line.
On the other hand, in Argentina there is no planning. The agents in the market have the expansion decision.
Projects are discussed in public hearings and the veto of 30% of the agents a¤ectd by the project block the
construction. Also, the project needs to be approved by the national regulatory agent.
França and Ramos (1998) a¢ rm that is in process a trend to transfer to the market the responsibility for the
improvements in the transmission system. Another characteristic present in many models is the indicative planning
development. This planning is a key piece, generally prepapred by government institutions, that aims indicate to
the market agents the optimal evolution in the generation and the transmission in the long term. With the planning,
the agents’decisions are oriented in terms of investments and contractual negotiatitons.
Concerning the tari¤ regulation, in the transmission systems in the United States it is mainly by rate-of-return
or cost of service regulation. Therefore, the transmission investors do not internalize the congestions costs in the
system, since these costs are repassed to consumers and, then, there is no incentive to investments in the reduction of
connected, as they …nally were later on. The conclusion is that the created lines after the 2001 crisis added about 2 thousand MW to
the SIN, without the necessity to build a single barrier, avoiding an estimated expense of R$2 billion.
5
the congestion cost5 . As a consequence, rate-of-return regulation may be inconsistent with newer forms of regulation
that seek to emulate the role of competitive market forces in eliciting e¢ cient behavior from regulated …rms. A
basic tenet of competitive markets is that investors are rewarded based on the value and innovativeness of their
actions (not on the cost of their investments, which is the basis for rewards under rate-of-return regulation).
A new class of regulatory approaches, called performancebased regulation (PBR), o¤ers greater promise in
o¤ering incentives toward this end. In England, for example, they have a performance based regulation that gives
incentives in the cost reduction and in investments in new transmission lines, with technology innovations. The
National Grid Company’s (NGC) PBR mechanism employs a pro…t-sharing approach to reward NGC for reducing
the charges that are passed on to consumers for recovery of congestion relief costs incurred by NGC. The pro…tsharing scheme is based on NGC’s performance relative to a predetermined “yardstick”set by the regulator in view
of historical performance and expected e¢ ciency improvements. NGC has reduced the costs of congestion through
a combination of operational e¢ ciency improvements, improved forecasting, investment in transmission expansion,
and adoption of technologies that improve transmission grid utilization.
An important comparison that can be made in terms of investments in the transmission systems is about the
choice between nodal pricing and zonal pricing. In this section we will compare a system that uses the nodal pricing
- the PJM system - with another one that uses zonal pricing - the Brazilian system.
Boucher et ali (1999) say that nodal pricing is meant to send generators the right investment and operations
signal. In principle, this is achieved when nodal prices properly re‡ect network congestion. Improving operations by
relying on accurate price signals may, by itself, alleviate the need for some construction of new transmission facilities.
Moreover, when new construction is needed, price signals will help market participants identify opportunities and
assess options to address bottlenecks. However, nodal pricing is not always well accepted in practice. It is commonly
argued that congestion only occurs infrequently and, hence, that nodal pricing is irrelevant. Furthermore, nodal
prices do not ensure the grid operator with su¢ cient revenues to recover the …xed charges of the network, requiring
that additional price components be added for that purpose.
The iconic model of economics-driven capacity expansion using nodal pricing is the PJM6 Model. In this context,
the ISO operates an energy spot market where locational marginal energy prices are based on a security-constrained,
bid-based, optimal dispatch. Congestion charges for a point-to-point transaction are priced at the opportunity costs
given by the locational price di¤erence between the two points.
Since 1998, a very interesting point in the PJM is the existence of hedge against the congestion costs volatility.
This hedge is done by the Financial Transmission Rights (FTRs), which take the form of …nancial instruments
that entitle their holders to the locational price di¤erence times the number of rights (in MW units) over the
speci…ed time interval. This instrument is equivalent to a physical right because it enables its holder to execute
a point-to-point transaction and o¤set the congestion charges with the FTR revenues. FTRs are auctioned o¤ by
5 In principle, any congestion problem has a “generation solution”, which amounts to creating counter‡ow on the congested interface,
and a “wires solution,” which requires investment in transmission assets. From an economic perspective, the optimal amount of
transmission capacity is achieved when the marginal cost of the generation solution and that of the wire solution are equal. This
equality represents the optimum solution from a social perspective, i.e., the total of consumer and producer surplus is maximized;
however, there is no guarantee that the consumer surplus increases at this solution. Transmission capacity mitigates market power, so it
is possible that additional capacity may bene…t consumers by facilitating trading and reducing energy prices although such investment
need not be optimal from a total welfare perspective. It is also possible that a transmission expansion that is socially desirable may
disadvantage some consumers by increasing their energy costs.
6 PJM is an American Regional Transmission Organization (RTO) which is part of the Eastern Interconnection grid operating an
electric transmission system serving all or parts of Delaware, Illinois, Indiana, Kentucky, Maryland, Michigan, New Jersey, North
Carolina, Ohio, Pennsylvania, Tennessee, Virginia, West Virginia and the District of Columbia. PJM is currently the world’s largest
competitive wholesale electricity market.
6
the ISO periodically for di¤erent time horizons. The FTRs that are issued must satisfy simultaneous feasibility
conditions, which require that if all FTR holders were to use their rights by scheduling corresponding transactions,
these transactions would be feasible without impacting the security of the system. This simultaneous feasibility
condition guarantees that the congestion revenue collected by the ISO can cover the FTR settlements.
As previously argued, Financial Transmission Rights are …nancial instruments that entitle to the winners of
the FTRs auctions to a stream of revenues (or charges) based on the hourly congestion price di¤erences across a
transmission path in the Day-Ahead Energy Market. FTRs allow transmission customers to protect themselves
against the risk of congestion cost increases, providing price certainty for transmission customers when receiving
electricity across constrained transmission lines. Transmission customers pay the market price to move energy
from one point in the transmission system to another. This cost increases when the delivery is across transmission
facilities that are constrained. FTRs are a necessary hedging mechanism in the locational pricing market.
Turning attention to zonal pricing in Brazil, Maurer (2005) say that nowadays Brazilian electricity sector is
operating in a tight pool model, with a centralized dispatch, looking for the smallest operational cost. This is an
international consolidated format, once it allow the best use of the generation and transmission resources. The
ONS calculate the spot price (shadow), based in the water value, the declared costs, the o¤ers received and, then,
minimizes costs taking into account all the transmission constraints. For contract liquidations, zonal pricing is
adopted, with four submarkets. The zonal prices do not take into account the constraints inside the submarkets.
The zonal pricing in Brazil was bring from the British model. In that time, 1997, the zonal pricing was seeing
as the better solution to the commitment between alocative signs and the necessity to avoid price volatility (vis-avis the nodal system). Another reason to the zonal pricing choice is that, in the occasion, the systems had little
experience in nodal pricing. Some disadvantage include:
(i) Agents would have to much exposure to price volatility;
(ii) Complexity.
(iii) Questionable value, once in most cases the constraints are not present;
(iv) Make the transactions di¢ cult and give small liquidity;
(v) Concentrate power market by disintegrating zones.
According to Maurer (2005), in the last 10 years these premises were seeing as myth, corroborating the nodal
model as the correct technical model. The mainly counterarguments are:
(i) Risk reduction could be made by congestion contracts, as the FTRs in the PJM model.
(ii) The complexity is intrinsic to the centralized dispactch, which has to calculate the marginal cost by node.
The accrual by zone and the arbitrary criteria to make it that bring further complexity to the system.
(iii) The price system need to capture the constraints when they exist. The zonal pricing system can give a
reasonable sign most of the time, but it does not work exactly when it is more necessary.
(iv) The liquidity objective can not be reached sharing costs. The international experience has demonstrate that
agents learn to deal with FTRs very easily, reaching the necessary liquidity in the market.
(v) Do not increase the market power, even when the plant is beyond transmission constraints.
Finally, the author say that regions where the nodal pricing, or LMP, was applied have a successful market, while
problematic markets applied zonal pricing. Among successful markets we can highlight New York, New England,
PJM and more recently Texas. On the other hand, among problematic markets is California, whose failure of the
system was predicted three years before the crisis, but not much was done to correct it.
This is an important question to Brazil especially to an adequate location of generation plants. If we do not have
7
constraints pricing inside the submarkets, the investor becomes insensible in terms of the plants optimal location.
However, we need to understand that the Brazilian model to capture transmission investments, with a guaranteed
revenue (regardless the tari¤ system do not give locational signs as it should be doing) was important in the context
of a new regulation in the energy electric sector. The improvements in the transmission was indispensable in a
system that present a lack of generation culminating in the 2001 Rationing Crisis.
4
Transmission Auction: Contractual Aspects of the Object of the
Study
The Government is the Granting Authority in the transmission service supply. The regulatory agency, ANEEL, is
the responsible agent for the auction accomplishment. After the evaluation stating which lines will be auctioned,
the edict is published, which contains information about the auction and the transmission line. For example, the
edict contains the main line connection points, precisely its origin and ending point, the line tension, its extension,
the order that objects will be auctioned, information for prequali…cation and the contracts to be celebrated between
the winning proponent and the other agents - ANEEL, line users, etc.
According to ANEEL7 , it constitutes the object of the transmission auction the concession hiring for the transmission service supply, by the smaller proposed annual income, in an individual way for each lot, including the
construction, operation and maintenance of the transmission plants characterized by a thirty year period from the
signing date of the concerning concession contract. It is accepted proposals that do not exceed the limit values for
the allowed annual incomes.
Regarding to the lines, the transmission company must set them in, complying with the determination in the
applicable environmental legislation, adopting all the necessary measures next to the responsible agent for the
license acquisition, on its own risk, and accomplishing all the demanded rules. Besides the environmental licensing
agency’s requirements, the transmission company will have to establish compensatory measures, detailed in the
Environmental Basic Project presentation, of its responsibility, next to the competent agency, and yet submit itself
to the environmental agent’s requirements of the states where the transmission plants will be established.
Transmission lines shall start the commercial operation in the term, calculated from the signing date of the
respective concession contracts. The construction, operation and maintenance of the transmission lines is of exclusive responsibility of the transmission company and, therefore, it is responsible for getting …nancial resources,
to develop directly or to contract the services from third parties, to get materials and equipment to set apart or
to be substituted. Yet the transmission company is responsible for the plant integrity, and to agree upon speci…c
regulation established by ANEEL and upon net procedures, as well as the established conditions in the concession
contract and in the CPST (Transmission Service Contract).
Regarding the term to initiate operations, to take exception from the provided hypotheses in the legislation and
in the concession contract, it will not be considered by ANEEL any complaint from the transmission company that
is based on, for example:
a) if the available studies and projects are inadequate or inaccurate;
b) in the ignorance of the local conditions that in‡uence direct or indirectly the terms for the delivery of the
materials, work force and equipment, as well as the project and construction terms;
7 Edict
No 005-2006_1109.
8
c) in the weather conditions, excess of rain, geology, geotechnics, topography, roads, regional infrastructure,
communication media, sanitary conditions and environmental pollution.
Besides that, the transmission company, in the same date or until thirty days after the celebration of the
concession contract, shall sign the CPST (Transmission Service Contract) with the System National Operator
(ONS), consolidate the technical and commercial conditions related to the transmission plant availability for the
interlinked operation.
Regarding to the transmission line utilization, the free access to the transmission plants is assured by the law of
2005 and regulated by ANEEL, and the transmission company shall sign the respective CCT (Transmission System
Connection Contract) with users in the limitations of the applicable law.
Inside the actual Interconnected System of the Brazilian Electric Sector, whose aim is the e¢ ciency in the
operations, the transmission company, complying with the function of the interconnected system and to allow the
access to its transmission lines by other transmission concessionaries, according to the regulation, must: make the
studies available, as well as the projects and standards used in its lines; grant a license or transfer, with previous
approval by ANEEL, possessions and necessary lines, with the aim of optimizing investments and to point out
in a better way the responsibilities by the transmission service supply; share lines, infrastructure and allow the
construction in available areas; and enter into the CCI (Facility Sharing Contract).
The allowed annual income of the transmission company responsible for the transmission service supply for
the …rst …fteen years of transmission plant availability for commercial operation will be the value of the winning
bidding proposal of the auction. From the 16th year of availability of the transmission lines for the commercial
operation, and until the end of the concession term, by the contracted service supply, the allowed annual income of
the transmission company will be of 50% of the allowed annual income of the 15th year.8
For the de…nition of the allowed annual income (RAP) for each lot of the various auctions, the parameters of
the table below are used, whose values vary from line to line. These same parameters constitute the base for the
Periodic Tari¤ Review provided in the concession contracts.
8 RAP
is readjusted annually by IPCA. The only discount is for line unavailability.
9
Table 1 - Tari¤ Periodic Revision Rules
Item
Parameters
Status to Periodic Revision
1
Equity Structure
2
Debt Structure
3
Real Cost of equity
4
Real cost of debt
Atualized at the moment of the
4.1
TJLP (Long-Term Interest Rate)
periodic reviews in terms of the
4.2
IPCA (In‡ation Index)
4.3
NTN-B
4.4
Spread s1
4.5
Spread s2
5
Operation and Maintenance
6
Depreciation Anual Mean Rate
Fixed for the Review stipulated
in the concession contract.
contract.
Fixed for the Review stipulated
in the concession contract.
* The values are indicated in the transmission lines contracts.
Source: ANEEL
From the above data, through the method of Discounted Cash Flow, ANEEL calculates the annual payment
series that pays o¤ investments associated to the transmission line in question.
If there is change in the costs of the concessionary, the RAP value of the contract can be reviewed, which makes
the transmission concession contracts an asset quite secure, with assured income, where the risk is pass on to the
consumers.
According to Hirota (2006), a criticism to this model is about the lack of incentives of the company to the service
universalization, because with the service ampli…cation, the company’s predetermined income will not increase.
In this way, ANEEL is responsible for the election of the places where the concessionary will invest, and the
concessionary is responsible for making the service available with the quality agreed.
The payment of the allowed annual income of the transmission company will be in twelve monthly quotas in the
terms of the concession contract and established in the CPST (Transmission Service Contract) and in the CUST
(Transmission System Usage Contract).
The prerequisites for the companies’ quali…cation that intend to participate in the transmission auction are
countless, as for example:
a) National companies that do not have been constituted with the speci…c purpose to explore transmission service
concessions, and the foreigners, interested in participating in the auction, shall present an obligation to constitute
themselves as a society with the speci…c purpose according to the Brazilian laws, established and administrated in
the country, to explore each concession of the transmission service to be contracted.
b) It is not allowed the participation, separately or in partnership, in the prequali…cation for a same lot, by a
company that integrates an economic group that is a proponent of this same lot.
c) The partnership participation will be allowed by means of a presentation of the partnership constitution
contract by a public or particular notary, subscribed by the partnership companies’legal representatives, of which
shall be discriminated in speci…c clauses.
10
d) In the partnerships constituted by the Brazilian and foreigner companies, the society leadership will always
be controlled by the Brazilian company.
After these stages, the responsible commission, with CBLC (Brazilian Settlement and Custody Company)
and Bovespa support, proceeds to the analysis of the prequali…ed documents, prepares the analysis report of the
documentation and draw up the minutes its decision. The commission publishes the decision of the prequali…ed
stage in the "Diário O…cial da União" (equivalent to the USA Federal Register) and also in the ANEEL website,
for public knowing 9 .
In the signing stage of the concession contract, the winning proponent of each lot shall present to the ANEEL,
in the term indicated in the schedule, the transmission lines budgets of each lot and the construction schedules of
the transmission lines of each lot. The presented budgets constitute, uniquely and exclusively, a reference used by
the transmission company to present the necessary values to the complete establishment of the transmission lines.
It does not represent, and it shall not represent, warranty as a base of capital remuneration or gain of any kind;
they shall not be used, in any circumstance, as a parameter for any lawsuit related to the economic and …nancial
balance maintenance of the concession contract and they shall not be used for application of the depreciation taxes.
The signing of the concession contract imposes to the transmission company the obligations and duties related
to the transmission service supply which shall be executed with the following characteristics: regularity, continuity,
e¢ ciency, security, generality, courtesy in the consumer service and tax moderation, according to the speci…c
legislation. The transmission company shall maintain, permanently and during the concession period, the technical
quali…cation equal or superior to that required and presented in the prequali…cation.
5
The Auction Rules
The con…guration of the transmission line auctions is a hybrid auction between two models - a sealed …rst price
auction and an English auction. In the …rst stage, each competitor makes a sealed bid. With the envelope opening,
if there are no close bids, the competitor is declared as the winner. If there is at least a bid near enough to the
higher bid, there will be a second stage in which the remaining competitors dispute in an outcry auction, where the
higher bid of the …rst stage is the reserve price.
This hybrid auction con…guration10 was proposed for the …rst time by Paul Klemperer (2002), for the execution
of the spectrum auction of third generation telecom in England. Klemperer states that the superiority of the model
is because it combines the positive characteristics of both isolated con…gurations. For example, he mentions that
the English model is the ideal way to allocate a commodity to someone who gives a real value to it, because a
competitor can exceed someone’s bid in any moment, till the good is acquired by that higher value. However, the
English model is subject to the predatory behavior, once the open bid allows retaliation to someone that does not
accomplish an agreement previously signed. In the case of the sealed model, its …rst positive feature is the absence
of the retaliation tool and a larger encouragement of the participants’ entrance, once the result is less uncertain
than that in an open auction. As a negative point, it is pointed out that an auction can be ine¢ cient because it
can fail to allocate the commodity to whom that puts more value to it.
9 In
the address http://www.aneel.gov.br, by accessing the menu "Licitações" ("Biddings") and submenu "Editais de Transmissão"
("Transmission Edicts") and the entailed documents , or in the Bidding Secretary, situated in the SGAN 603, Módulo J, sala 211, Asa
Norte, Brasília-DF, with the reasons of the eventual disquali…cations.
1 0 Actually, only with the stages changed, that is, with the open stage preceding the sealed stage.
11
In this scenario, hybrid auctions have become a mechanism more and more used. Besides to combine the
positive features of both con…gurations described previously, there is another characteristic of the hybrid auctions
very interesting to the granting authority and to the society. Dutra & Menezes (2003) proved that the hybrid
con…guration above, also used in the telecom privatization process in Brazil, gets more income to the auctioneer
than any other auction con…guration:
"The …rst price auction with a Vickrey auction as a second stage when there are bids su…ciently close
to the top bid guarantees a higher expected revenue to the seller as compared to standard auction
mechanisms."
In technical terms, the auction is divided in lots, each one comprising a hiring of the concession for the transmission service supply of electric energy, including the construction, operation and maintenance of the transmission
plants of the basic net of the National Interconnected System. The auction purpose is to grant the public service
concession of these lots by the best possible o¤er, according to the described proceedings in this chapter. The bids
are presented in a closed envelope with an open outcry auction.
The summary of the main characteristics of the mentioned auction is listed below:
(i) Proponent’s access:
Only the quali…ed proponents that are in the announced list by the commission in advance, will have access to
the auction, subordinated to the presented warranties.
(ii) Representation by the broker:
Each broker shall represent only one proponent by each lot, inasmuch as the proponent is quali…ed in each lot.
(iii) Composition of the proponents or partnership:
Each proponent shall only have one composition for each lot in the auction.
(iv) Objective:
Concession hiring for the transmission service supply.
(v) Allowed Annual Income:
It will not be accepted proposals that o¤er to the Allowed Annual Income values higher than an established
speci…c amount by each lot in question.
(vi) Cuto¤ value:
It is calculated over the value of the smallest proposal in the envelopes, according to the following formula:
V C = LV x 1; 05
(1)
Where, VC = Cuto¤ Value;
LV = Value of the smallest proposal in the envelopes.
(vii) Winner Announcement in the Auction by each Envelope:
The winner of each lot is the proponent who presents the bid of smallest value in the envelopes, in the event of
the other bids are higher than the cuto¤ value.
(viii) Outcry auction:
If there are more bids with a smaller or equal value than the cuto¤ value, the auction continues by successive
bids made by the outcry auction. The outcry bids are made by the representative brokers of the proponents. The
proponent that has presented the smallest bid in the envelope and those whose o¤ers in the envelopes have been equal
12
or smaller than the cuto¤ value can participate in the outcry auction. The director of the outcry auction, before its
beginning, informs the value of the smallest bid. From that value, the brokers that participate in the outcry auction
shall proclaim a new price, smaller than the price proclaimed by the others. The prices are proclaimed in "Reais".
The price variation takes place in free intervals, and the director of the auction can, however, during its course,
establish an interval or change it to less or more, if he judges this providence necessary to the good proceedings of
the auction. Each bid is considered as a strong o¤er, must the proponent to honor it unconditionally, if the auction
is closed.
(ix) Winner Announcement:
The winner of the auction is that who presents in the outcry auction the smallest bid. If there is not an outcry
bid, it is announced as a winner the smallest bid inside the envelopes. If there is more than one envelope with the
smallest price and if there is not an outcry bid, the director of the auction proceeds with the lottery between the
proponents with the same classi…cation to choose the winner.
As each envelope is open, the director of the auction reads the bid value, which is the Allowed Annual Income.
When the bid is read, the director of the auction waits for the typing of this information in the Bovespa auction
system and their con…rmation. Once the bid is con…rmed, the director of the auction proceeds to the open of a new
envelope.
(x) Bid Con…rmation:
The con…rmation of each bid is dependent upon:
a) the bid value is equal or smaller than the Allowed Annual Income;
b) the proponent’s warranty value is equal or higher than a minimum established by each lot.
The inability to accomplish to any of the previous conditions causes the proponent’s disquali…cation. Once these
proceedings are accomplished, the winner of each lot will be announced.
6
Literature Review
The empirical study of the auctions is more and more common in the economic science, not only because its
political practical appeal, but also because their data are, in many times, better than that typical data of industrial
organization. This is due to the following features:
(i) the game is relatively simple, with well speci…c rules;
(ii) the participant’s actions are observed directly;
(iii) payo¤s can be deduced.
According to Wolfram (1998), the existent empirical literature about auction is extensive, presumably due to the
information richness related to the auctions and to the extensive theoretical treatment dedicated to them. Empirical
papers analyzed markets like concessions for oil, wood, and aubergine exploration. With the exception of many
auction studies for …nancial instruments and auction analyses of the American radio spectrum (FCC), the great
part of the empirical literature takes into consideration the auctions where the participants compete for only one
object.
Hendricks & Porter (1988) compare the normative and positive objectives of the auction theory.
13
Table 2 - Positive and Normative Aspects from Auctions Theory
Positive Theory of Auctions
Normative Theory of Auctions
It describes how one can o¤er a bid
It seeks to characterize the
rationally, which normally comprehends
best or e¢ cient sale mechanism.
to characterize the Nash Bayesian
Equilibrium (NBE) of the game.
In the same way, an empirical study about auction also has its positive and normative objectives, as below:
Table 3 - Positive and Normative Aspects from Empirical Auctions Works
Positive View - Empirical Studies of Auctions
Normative View - Empirical Studies of Auctions
To answer to the questions about how
To answer what is the maximum income
the agents behave themselves and if the
or the e¢ cient auction.
values that they attribute to the object
are correlated. If so, what is the source of
correlation? The observed distribution of
the bids is consistent with the NBE?
The aversion to the risk is relevant?
According to Athey & Haile (2001), the standard practice in the empirical literature of auctions is considering
a model of strategic demand and information, as the model of independent private values, based on the qualitative
evidence in spite of the application. As di¤erent hypothesis can get to results and implication of policy too di¤erent,
a formal base to evaluate alternative models could be preferable and it would take to more trustful empirical results.
Bajari (2000) stands out a series of di¢ culties in the econometric analysis of auctions, which are:
(i) auctions are games with functions of discontinued utilities;
(ii) existence, uniqueness and characterization are established for a restrict set of hypothesis;
(iii) even when these basic results are valid, it is di¢ cult to calculate the equilibrium of the auction models;
(iv) if the equilibrium is not unique and if it cannot be calculated e¢ ciently, it is not possible to estimate the
parameters using a verisimilar function.
(v) situations where the equilibrium exists in an auction game and whose equilibrium is unique are very limitated.
The auction analysis can follow estimation in two con…gurations: structural and reduced. Bajari (2000) points
out that tests of the auction theory using reduced con…gurations to analyze the implication of the strategic behavior
have a poor statistic power. On the contrary, structural models allow to shape the correlation in the informational
structure, to compare the alternative auction con…gurations and can be used to deduce a collusive behavior in
biddings.
Bajari & Hortaçsu (2005) state that the structural con…guration is based on three strong hypotheses:
(i) the competitor’s objective is to maximize its expected utility;
(ii) competitors are able to calculate the relation between their bids and the probability to win the auction;
14
(iii) because their believes, competitors are able to maximize correctly their expected utility.
To Shum & Hong (2000), the two approaches - structural against reduced estimation - are used to answer the
di¤erent kinds of questions. The reduced con…guration is more used to characterize a behavior of the player’s bid in
an auction than to list the bid functions of equilibrium from the observed bids in order to estimate the distribution
parameters.
Rezende (2005) points out that the empirical methods created to investigate auction data are interesting, not
only because they allow testing the auction theory, but also because they provide a rich source of information
about consumers’preferences or production costs who participate in auctions. To this purpose, it would be valid to
use a based empirical strategy, strong against failures of speci…cation in the environment details and, also, of easy
implementation. The author suggests this method.
Studies that usually try to understand how the covariables a¤ect the demand in the auctions estimate regressions
like:
p=X +"
(2)
where p is the transaction price (or its logarithm). The problem with this analysis does not come from the right
side of the equation - covariables can, in fact, impact the demand in a linear way - but at the left side. An initial
point for a superior empirical analysis would be to use the following speci…cation:
Vi = X + "
(3)
where Vi is the value that the n-th consumer attributes to the commodity i, or the availability to pay by
the product that is being auctioned. Di¤erently from the price, this value is a demand concept that re‡ects the
consumer’s preferences in an isolated way from the o¤er e¤ects or from the market institutions. A consumer will
never choose to pay exactly the real value estimated by him for an object, because in this way he will not take an
advantage in participating in the auction. Thus, price and value usually have distinctive magnitudes, making the
estimation of the …rst one of these two regressions from above undergoes from speci…cation bias.
As it was observed before, reduced models have a poor statistical power to analyze the competitors’behavior
implication. On the other side, Rezende(2005) states that structural methods used nowadays present two weak
points regarding the applicability:
(i) They are designed for a speci…c type of auction;
(ii) They are complex in terms of calculation.
In this way, the author presents a new model that, being structural, is justi…ed theoretically. Besides that, it is of
simple calculation and supports the many types of bad speci…cations regarding to the auction that is being played.
The proposed method comprehends a regression of the transaction prices observed (winning bids) by minimum
norm least squares in the covariables of interest and in an additional regressor with the number of competitors
in each auction. This method is related to the other exercises yet accomplished in the empirical literature (e.g.
Gilley and Karels,1981) either in the reduced con…guration, or in the structural (e.g. Paarsch, 1991). The great
di¤erence is that the proposed model by Rezende (2005) inserts these competitors in a nonlinear way, in order to
justify theoretically the regression method11 .
1 1 The
author proves that hypotheses of linearity or in a quadratic fashion for the number of competitors are not correct.
15
This model explores two auction properties - the Revenue Equivalence Theorem and the equilibrium set invariance to the equilibrium set to the related transformations in the attributed value to the object by competitors.
Both were approached in other models, but they had not been still considered together and, thus, the previous
methods were very demanding in terms of calculation or they were applicable to a limited class of auctions.
According to Rezende (2005), the estimation of the proposed model can be made by two ways. In the present
paper, we will be restricted to the second mode of estimation - which prevents the investigator from imposing a
con…guration for the distribution of the attributed values to the commodity by competitors.
Once we have a well de…ned institutional environment, it is possible to use the auction theory to get a mapping
between the values and the behavior of the observed bids and, from this, to develop a strategy of estimation. La¤ont
& Vuong (1996) showed that, depending on the auction protocol, the distribution of the values is readily identi…ed
by the observed data. Many structural methods are used to estimate the values from the auction data, but they
are made for a speci…c type of auction, besides being complex in terms of calculation.
Rezende (2005), thus, suggests a new method in question which involves the estimation of a regression of
minimum norm least squares of the transaction prices observed in the covariables of interest and an additional
regressor with the number of competitors in each auction.
Many empirical papers apply the auction analysis techniques described above. We will see three of them in a
closer way, which are the base for the utilization of some variables in the present paper. They are the papers from
Jofre-Bonet & Pesendorfer (2000), Bajari & Ye (2001) and De Silva (2005).
Jofre-Bonet & Pesendorfer (2000) studied data from highway biddings for the state of California, from December
1994 to October 1998. To this study, it is used a model of dynamic auctions that takes into consideration the
existence of intertemporal restrictions as that of capacity. Using nonparametric statistics this model is estimated
and reveals the presence of dynamic restrictions in the o¤ers.
Data from highway construction have been studied by many authors. Feinstein et al. (1985) and Porter & Zona
(1993) studied aspects related to the collusion. Bajari (1997) focused in the importance of the asymmetry between
players.
Jofre-Bonet & Pesendorfer’s model has the following description:
In…nite periods of time. In each period, the buyer o¤ers a unique contract to the sale. There are two kinds of
competitors: regular and outliers. The …rst ones remain in the game forever and the others have a short life
and abandon the game in a determined period;
In each period the competitor i learns about his cost for the contract. The cost is a private information and
denoted by cit . The belief of all competitors about the cost cit from a regular competitor is identical and
represented by a distribution function F (cjsit ; sot ), where sit is the variable that summarizes the available
capacity to the competitor i in the time t; and s t denotes the project characteristics. In the same way, the
competitor outlier’s cost comes from a distribution function F (cjsot ):
Neutral agents to the risk. The evolution of the state will depend on the amount of projects and on the size
of the project won in this period.
They are considered perfect balances in subgames and markovian strategies;
A strategy for i is a function of the state vector s and of the cost of the competitor i in the period of time t.
16
And the authors’econometric estimation has the following stages:
(i) They show how is made the choice of the state variables;
(ii) They describe the estimation of the bid distribution function, G, that represents the competitors’belief;
(iii) Finally, they explain how to use G to deduce the costs and the cost distribution function.
At …rst, the bid log is projected in a vector of the characteristics of the contract and the density and distribution
of the residues b are estimated;
In the second place, the bid distribution function is estimated using the following variables:(ai ; a i ; r ; x i ; W i );
ai } is the backlog of i ; r and xi are the indexes that measure the competitor’s capacity and localization characteristics (the …rst one measures the number of competitors who are next in the place of the contract and the second
index is a linear projection of the decision about submitting a bid in the period of the maximum capacity of the
company. W i measures the number of contracts won by i before t).
Optimum bids are chosen based on the private costs, in the belief of the bids made by the other competitors
and in the e¤ect of all these factors on the future payo¤s. The necessary …rst-order condition for the bid balance
summarizes this relation and it is from it that the costs are deduced. They chose a nonparametric method to
estimate the model in the paper because the authors did not know about the con…guration of the competitors’
belief.
In conclusion, the authors state that the behavior of the bids is a¤ected by the restriction of capacity. Competitors with a higher amount of work from old projects have, in average, a higher cost in relation to the competitors
with low amount.
Bajari & Ye (2001) develop an auction model where the asymmetry between competitors is introduced. In the
paper, it shows how the asymmetries can appear due to the localization, capacity restrictions, and collusion between
competitors. The authors also demonstrate how to test the conditions that characterize a competitive environment
applying the test to a data set of bidding contracts.
In Bajari & Ye’s model (2001), competitors provide sealed bids and the contract is given to the agent who has
provided the smaller o¤er. Before starting the game, each company makes a cost estimation for the project, which
is a private information. Companies are risk-neutral and have private values.
This model is made taking in consideration that, in many bidding processes, asymmetries arise between competitors due to the localization (companies near the construction site of the project will be more capable to o¤er a
bid for this project due to smaller costs of transportation) and due to the capacity utilization (some companies are
small regarding to the market and need to take in consideration the fact that they won a project before participating
in others). Technological di¤erences can be a source of asymmetry between the companies and they can have a
bigger or smaller chance to win a contract depending on the State/Country that they are more active (companies
need to become familiar with regulation and policies of the local bidding).
In technical terms:
(i) In the …rst place the model information structure is described.
N …rms compet for a contract to construct a project. An estimate of each …rm cost is a random variable Ci
with realization ci: A random variable Ci has accumulated distribution function Fi (:; i ) in wich i is a parameters
vector. The cost has support c = (c; c) for all …rms. Consider an example in which each …rm has a location and,
therefore, a transportation cost. One especi…cation for the private information of this …rm is:
ci = cons tan t + 1 :dis tan cei + "i
17
(4)
In the equation, the …rm’s cost estimate ci is a function of the constant term –that can re‡ect attributes of the
project that a¤ect all …rms identically, as the road-paving needs or how many concrete tons gonna be used for a
building base. The second term is di¤erent for each …rm, because each one has its own location; the third and last
term, a random variable, is used to model the private information about the …rm’s cost, as the material or work
cost. If "i is normally distributed with zero mean and standard-deviation , so i = (constant; 1 ; distance; ):
(ii) After, the Expected Utility Function for the …rms is de…ned;
Let bi be the …rm’s i bid. If …rm i submets the smallest bid, your utility vNM is bi c i ;other way around, is
zero. In the collusion context, it’s possible to see that the cartel behaviour working e¢ ciently is a simple special case
of the model just exposed. Suppose that, before the auction starts, every cartel participant do their cost estimates.
The cartel members meet before the auction, compare cost draw and the cartel member with the smallest estimate
submits an actual bid, while the others choose not to bid or submit fake bids. Let C f1; 2; :::; N g be the cartel.
The cost to the cartel that we denote by cc can be writing as:
cc = min cj
(5)
j2C
If others competitors know the cartel members, so is trivial to adopt the precedent model for the cartel case.
The cartel is simply modeled as an order statistic of the costs estimates of their members.
(iii) Finally, the equilibrium strategic function is characterized.
In the model of asymmetric auctions, we take that …rms have a Bayesian-Nash equilibrium with pure strategies.
The i …rm, …rstly, make a cost estimate ci , and so, given that cost, choose a bid that maximizes its expected pro…t:
i (bi ; ci ; b i ;
) = (bi
ci )Qi (bi ; )
In which Qi is the probability that i …rm give the smallest bid.
O modelo de equilíbrio pode ser caracterizado como solução de um sistema de equações diferenciais com condições
de regularidade, quais sejam:
For each i , the cost distribution function Fi (ci ; i ) has support [c; c]. The probability density function is
continually diferenciable in c.
For each i , both fi (ci ; ) are restricted by zero in [c; c].
The model is, therefore, a simple variation of the "Independent-Private Value Model”, that allows …rms to be
asymmetric.
All things considered, is important, in this literature analysis, emphasizes previous papers that pursuited to
identify the e¤ects of possible synergies in the strategic behaviour of competitors.
De Silva (2005), for instance, examines the impact of synergies in the competitors behaviour in road construction
auctions. The article revels that projects are spacially correlated. So, his conslusion is that, when a competitor
with a potential synergy enters into the auction, the results show that the winning probability of this competitor
increases, once his bid is more agressive. Finally, the paper also shows that …rms without capacity constraints have
a more agressive bid compared with the …rms with this kind of restraint.
In accordance with De Silva (2005), empirical papers about auctions synergies are scanty. Gandal (1997) showed
that complementarities associated with the achievement of multiple projects in a particular geographic area improved
18
the value of TV cab licenses close one to another in Israel. Ausubel et al. (1997), for his turn, indicated that may
be synergies in winning adjoining licenses in the American communication auctions. Rusco & Walls (1999) show
that, in repeated wood auctions in correlated spaces, competitors with the possibility to have complementarities
with more than one project practiced more agressive bids.
Besides that, De Silva (2005) highlights that, when we looked for a possible strategic behaviour in bids on account
of synergies between auctioned objects, if we do not consider in our estimation that bids are di¤erent between
competitors and projects because of asymmetric costs - transportation or work cost -, only look for synergies,
without control for these variables, make the researcher to catch this information in the residuals.
Our results are complimentary to the others realized about Brazilian transmission auctions - Bueno & De Castro
(2006a) e Hirota (2006).
7
Empirical Exercises
The …rst objective of the present empirical exercise was model the winnig bid of the transmission lines auction in to
the factors that the literature considers relevant to the strategic behaviour analysis of multiple-object auctions bids.
The line we follow was the one given for the articles we presented in the last section: Rezende (2005), Jofre-Bonet
& Pesendorfer (2000), Bajari & Ye (2001) and De Silva(2005), looking for test reduced-based models that includes
the variables considered in these papers, that worked with similar auctions.
7.1
Data
We utilized the Brazilian data from all auction occurred from 2000 till middle of 2008, when were e¤ectively conceded
81 transmission lines (or lots of lines). 12 The source of our data was the Brazilian electricity Regulatory Agency
(ANEEL).
Independent Variables
Dependent Variables
Table 4 - Variables
Variables
Description of the Variables
BID
Winning Bid for each transmission line or bundle of lines
LBID
Log of Winning Bid
BID / INVEST
Fraction between Winning Bid and Estimate Investment
EXT
Extension of each line
COMP
Effective number of competitors in each auction
RAP(0)
Initial Anual Receipt Allowed
INVEST
Estimate Investment for Construction, Maintenance and Operation
TRIBUT
Dummy for winning bids bellow R$ 48 millions
EPC
Dummy for the presence of a “EPC”company in the winner union
SUL, SE,NE,CENO Dummies for regional location of the transmission lines
Dummies compet.
1 2 For
Dummies for number of competitors
the G lot of auction n o 005 of 2006, some data were not obtained, so thisobservation was excluded of our sample.
19
The following series are used as dependent variables in our reduced-based models: winning bid in each auction;
the log of the winning bid in each auction and the fraction of the winning bid and the estimated investment for the
project.
As independent variables, we included: the Allowed Initial Anual Revenue (RAP-0), that is equivalent to the
reserve price stipulated by the regulator agency; the number of competitors in each auction - a standard control
in the empirical auction literature; the estimated investment by the regulatory agency ANEEL for construction
and maintenaince of the lines13 ; extension of each line (factor that controls for the characteristics of the project);
dummy variable for winning union that has an "EPC" (Engineering, project and construction) in the group; dummy
variable for winning bids below R$ 48 millions.
Besides the varialbes mentioned before, we included too the hypothesis initially proposed that competitors that
already operate a transmission line will bargain more to operate another transmission line close to the …rst one,
that is, this company will o¤er a smaller anual revenue to win the second line because of the scale gains in operate
lines close one to another. To pursue this objective, we will add dummy variables for the lines location, as we saw
in De Silva (2005). We split the auctioned transmission lines into four macro-regions, evidentiated in …gure 2.14
Figure 2 - Brazilian Regions Speci…cation to Capture Synergies
According to Francellino & Polito (2007), in each transmission auction, agents surprise themselves with the
markdown o¤ered by Spanish transmission companies - Abengoa, Cymi, Isolux, Cobra e Elecnor ( the table 6
1 3 The
1 4 It
estimated investment for each prject it will be used as a proxy to the companies costs, in the lack of other variables.
was not possible to obtain property data of the transmission lines already operating before the implementation of the auctions
analysed. It would be a way to observe case-by-case which lines each company would seek in an evident way.
20
presents the winning concessionaries and the di¤erence between their winning bid and the reserve price in the
auctions analysed; in this table we don’t see the name of these Spanish transmission …rms, just the name of the
union, when it exists). In a speci…c event, the markdown o¤ered by these companies was more than 50% of the
reserve price stipulated. After this auction, many hypothesis has been formulated to explain the success of these
…rms. The authors say that there are two key aspects in the composition of this scenario:
(i) The fact of a competitor have or not have an "EPC" …rm15 in his group (with no exception, every Spanish …rm
that participated in the Brazilian transmission auctions have a company specialized in the EPC in their group). In
the bidding time, the EPC …rm pro…t margin get in the bid composition, making it more competitive than the bids
of other companies that don’t have an EPC. So, a dummy variable will be included to discriminate the competitors
with a construction …rm.
(ii) Many …rms o¤er a bid below R$ 48 millions - and, as a consequence, we see big di¤erences between the
winning bids and reserve price - to stay in a di¤erentiated tax system (the companies that register an annual revenue
below R$ 48 millions use a assumed pro…t income tax system). We includ this item as a dummy variable for bids
below this monetary value.
Transmission sector analysis (in Brasil Energia, 2007) assert that Spanish renounced a bigger internal rate of
return to pro…t just with the construction of the lines, and not with its operation and maintenance. Thus is expected
that projects with foreign …rms will be resaled in the future.16
Table 5 - Descriptive Statistics
Mean
Median
Maximum
Minimum
Std. Dev.
Skewness
Kurtosis
Observations
7.2
BID
29236.41
14376.00
140950.0
665.0000
32872.73
1.636670
4.989893
INVEST
221672.6
113651.3
994581.7
10479.00
234192.5
1.366879
3.902031
EXTENSION
307.6926
198.0000
1278.000
1.500000
280.4169
1.292265
4.165505
COMPET
5.987654
6.000000
14.00000
1.000000
3.540811
0.448843
2.235853
RAP_0
41759.95
24596.92
204902.1
1410.320
44301.30
1.475504
4.635172
BID/INVEST
0.131981
0.118778
0.226330
0.052523
0.049924
0.368450
1.862030
81
81
81
81
81
81
Reduced-Based Models Results
In this section we search to …nd what variables explain the winning bids in the Brazilian transmission auctions. As a
…rst step of a multivariate method, we test if all the potential explanative variables have the desirable characteristics
1 5 EPC
- Engineering Procurement and Construction.
feito pela ANEEL mostra que, do início dos leilões de transmissão até 2007, ocrorreram 21 alterações nas sociedades
de propósito especí…co constituídas para explorar as linhas de transmissão. Uma parcela dessas alterações foi apenas de empresas do
mesmo grupo societário. No entanto, outra parcela mostra que as linhas trocaram de "donos" algumas vezes.
1 6 Levantamento
21
for a linear model. We found that the Estimated Investment and the Reserve Price have a strong correlation. This
correlation between independent variables is a multicollinearity problem. According to Hair (2005), besides the
e¤ects in the explanation of our dependent variable, the multicollinearity can be serious e¤ects in the coe¢ cients’
regression and in the applicability of the estimated model.
Many possibilities are proposed in the situation if multicollinearity (Hair, 2005) , and the procedure more
employed is the deletion of one of the involved variables. For us, a more relevant information is the estimate of the
relationship between the winning bid and the investment variable (that will be our cost proxy), so we will drop out
the reserve price from these …rst reduced-based models.17
Figure 3 - Investment and Reserve Price Correlation
Invest and RAP_0 Correlation = 0.97
1,000,000
800,000
600,000
400,000
200,000
0
10
20
30
40
RAP_0
50
60
70
80
INVEST
Following we present the results from the four reduced-based models estimated for the winning bid transmission
auctions analysis. At this stage of our article, one can say that this empirical exercise could be desnecessary if one
prove that the high markdown observed was just a consequence of a change in the Regulatory Agency reserve price
formulation pattern. This was not the case, because we test the stability in the reserve price per kilometer. A same
pattern was found in our auction’s sample. So the pattern change in the winning bid level is due to something else
that we want to explain quantitatively by our models.
The …rst reduced-based model has this speci…cation18 :
BIDi =
1 EXTi + 2 COM Pi + 3 IN V
ESTi +
5 T AX+ 6 SOU T H+ 7 SE+ 8 N O+ 9 CW N O;
(6)
in wich,
BIDi : Winning bid in each auction i of transmission lines or bundle of lines;
1 7 However,
in the next section we will present an alternative solution to deal with multicollinearity without exclude any of our
informative variables. It is the Principal Component Analysis.
1 8 In this regression and in the next ones, we do not include the statistical non-signi…cant variables.
22
EXT i : Extension of each line or bundle of lines;
COM P i : Number of competitors in each auction i of transmission lines;
IN V EST i : Estimate Investment to construction and maintenance of transmission lines;
T AX : Dummy variable that indicates if the concessaionaire win the auction with a bid below R$ 48 millions;
SOU T H : Dummy variable indicating if the transmission line auctioned is mainly locate at South region;
SE : Dummy variable indicating if the transmission line auctioned is mainly locate at Southeast region;
N E : Dummy variable indicating if the transmission line auctioned is mainly locate at Northeasth region;
CW N O: Dummy variable indicating if the transmission line auctioned is mainly locate at Center-West or North regions.;
In this …rst model, with the winning bid as dependent variable, as we can see in table 7, we have:
The bigger the auctioned transmission line extension, bigger the winning bid;
The bigger the proxy to the line cost, given by the Estimated Investiment, bigger the winning bid;
The bigger the competitors number in the auction, smaller the winning bid;
The dummy variable that indicates that …rms gave bids below R$ 48 millions is signi…cative at 1% signi…cance
level, with a negative coe¢ cient, as we could expect;
Dummies indicating geographic location of auctioned transmission lines are signi…catives at 1% signi…cance
level.
23
Table 6 - Results of Estimated Reduced-Based Models
Variables1
Model 1
Number of Competitors
Extension
Coe¢ cient
t-Stats
-1585.1***
-5.08
41.22***
Tax Dummy
Model 2
Coe¢ cient
Model 3
t-Stats
Coe¢ cient
t-Stats
-0.01***
-10.16
2.89
-25889.8***
-3.84
0.04***
2.69
South Region
38642.8***
5.13
0.21***
Southeast Region
36498.7***
4.92
0.19***
22.18
Northeast Region
36226.9***
4.97
0.018***
17.69
North and Center-West Regions
34407.4***
4.75
0.19***
20.13
Estimated Investment
-0.17***
-3.02
Log Number of Competitors
-0.39***
-9.39
Log Estimated Investment
0.01***
63.69
Log Extension
0.02**
2.51
Model 4
Coe¢ cient
t-Stats
41.01***
2.85
-22779**
-3.05
0.05***
2.78
20.32
Dummy 1 competitor
30607.07***
3.23
Dummy 2 competitors
29923.89***
3.07
Dummy 3 competitors
25147.73***
2.80
Dummy 4 competitors
25421.69***
2.89
Dummy 5 competitors
20626.79*
1.81
Dummy 6 competitors
24048.61**
2.48
Dummy 7 competitors
16223.32*
1.67
Dummy 8 competitors
11336.19
1.13
Dummy 9 competitors
12603.15
1.19
Dummy 10 competitors
17090.75*
1.76
Dummy 11 competitors
17285.73**
2.03
Dummy 12 competitors
331.759
0.02
Dummy 13 competitors
19349.76**
2.28
16282.07*
1.85
Dummy 14 competitors
Adjusted R2
0.91
0.95
0.56
0.91
SQR
6.87E+09
5.28
0.08
5.90E+09
Schwarz Inf. Criterion
21.5
0.32
-3.72
21.9
Models 1 and 4 have the winning bid as dependent variable; model 2 has the log of the winning bid;
Model 3 has the fraction winning bid per estimated investment.
Signi…cance Levels: * 0,10 **0,05 ***0,01.
The second estimate follows this speci…cation:
LBIDi =
1 LCOM P i
+
2 LIN V
24
EST i +
3 LEXTi
+
4 T AX
(7)
In this regression, we have the logarithm of winning bid as dependent variable. We can conclude that:
The bigger the logarithm of competitors number, smaller the logarithm of the winning bid;
The bigger the logarithm of the line extension, bigger the logarithm of the winning bid;
The bigger the logarithm of the estimated investment, bigger the logarith of the winning bid;
The TAX dummy variable was signi…cative in this model, with a negative sign.
The third estimate, according to Bajari & Ye (2001), presents the following speci…cation, with the fraction
winning bid per estimate investment as dependent variable,
BID
=
IN V ESTi
1 COM P ET i
+
2 SOU T H
+
3 SE
+
4N E
+
5 CW N O
(8)
The results from this regression show these relations:
The bigger the competitors number in the auction, smaller the ratio winning bid per estimated investment;
All geographic dummies for transmission lines location are signi…cative at 1% signi…cance level.
At last, we estimate the regression:
BIDi =
1 EXT i + 2 IN V
EST i +
3 T AX
+
4 d1
+
5 d2
+ ::: +
17 d14
(9)
In which the dependent variable is the winning bid either, as in the …rst regression, but with the Rezende (2005)
approach. As we showed before, the author says that this estimation is a way to capture structural analysis into
the auctions econometrics. What Rezende does is a regression of the price - in our case, the winning bid - with the
covariables that a¤ect the mean price and in with the combinations of the covariables a¤ecting the variance with
the dummies for the number of competitors.
Then, in the speci…cation above, besides the already presented variables, we have the dummies with the number
of competitors in the auctions - d1 to d14 , once 1 to 14 are the possible values that the number of competitors can
assume in this analysis.
From the results of this exercise we can say that:
The bigger the line extension, bigger the winning bid;
The bigger the estimated investment, bigget the winning bid;
The dummy variable indicating that companies presented a bid below R$ 48 millions is signi…cative at 1% of
signi…cance level, with negative coe¢ cient, as we could expect;
Almost all the dummies for the number of competitors are signi…catives - eleven of these fourteen dummies
variables.
25
According to Rezende (2005), once we have estimated the regression above, we can use the parameters from the
dummies to …nd the competitors’ valuation distribution function. Our estimation, presented in …gure 4, suggest
that a better speci…cation for the competitors’valuation is given by the logistic distribution function.
Figure 4 - Competitors’Valuation Distribution Function
Empirical and Theoretical Distribution - Model 5
.00006
.00005
Density
.00004
.00003
.00002
.00001
.00000
-10,000
0
10,000 20,000 30,000 40,000 50,000
Logistic
Kernel
Another characteristic of the four reduced-based presented is that no one of them has the variable "EPC" - the
dummy for a construction …rm in the auction’s winning group - statistically signi…cative.
Graphs 5 to 8 show the estimated regressions, with the residuals of each one. We have some evidence, with the
…tting, that speci…cations with winning bid as dependent variable are more appropriate in the analysis. Then we
have …rst and fourth models explaining in a better way the bidding process in the transmission auctions.
26
Fig. 5 - Winning Bid Regression*
Fig. 6 - Log(winning bid) Regression
160 ,000
12
120 ,000
11
8 0 ,0 0 0
10
9
4 0 ,0 0 0
.6
3 0 ,0 0 0
8
0
.4
- 4 0 ,0 0 0
.2
2 0 ,0 0 0
7
1 0 ,0 0 0
6
.0
0
-.2
- 1 0 ,0 0 0
-.4
- 2 0 ,0 0 0
- 3 0 ,0 0 0
-.6
10
20
30
Residual
40
50
Actual
60
70
80
10
20
30
40
Residual
Fitted
Fig. 7 - Winning Bid/ Investment Regression
50
Actual
60
70
80
Fitted
Fig. 8 - Wining Bid Regression
.24
160 ,000
.20
120 ,000
.16
8 0 ,0 0 0
.12
.08
.08
.04
.04
.00
3 0 ,0 0 0
4 0 ,0 0 0
2 0 ,0 0 0
0
1 0 ,0 0 0
- 4 0 ,0 0 0
0
- .0 4
- 1 0 ,0 0 0
- .0 8
- 2 0 ,0 0 0
10
20
30
Residual
40
50
Actual
60
70
80
10
Fitted
20
30
Residual
40
50
Actual
60
70
80
Fitted
Graphs 9 to 12 below report the Jarque-Bera tests to the four regressions estimated in this article. Once the
null hypothesis of the Jarque-Bera test was not rejected in the tests we have normally distributed errors to these
equations. The aleatory term to be normaly distributed is important for the tests and fot the parameters estimation.
With normality, we can use F and t-Student tests to analyze the minimum square estimators we had found.19
1 9 Normality is justi…ed by the Central Limit Theorem. If error terms are identical and independently distributed with 0 mean and
variance 2 , estimators are assintotically normally distributed. Without normality, t and F tests are still valids to big samples, but not
to small ones.
27
Fig. 9 - Winning Bid Regression
Fig. 10 - Log(winning bid) Regression
12
9
Series: Residuals
Sample 1 81
Observations 81
10
Series: Residuals
Sample 1 81
Observations 81
8
7
Me a n
- 2 .7 8 e - 1 2
Me d ia n
1 5 9 .1 8 4 8
Ma x imu m
2 4 8 6 7 .1 1
Min imu m
- 2 3 5 7 4 .7 0
Std . D e v .
9 2 6 6 .8 0 7
Sk e w n e s s 0 .2 9 2 2 0 0
Ku r to s is
3 .4 6 2 4 8 5
8
6
4
J a r q u e - Be r a 1 .8 7 4 5 3 1
Pr o b a b ility 0 .3 9 1 6 9 7
2
5
4
3
2
-0.002716
-0.023218
0.4804 39
-0.515308
0.256890
-0.0020 17
2.102735
J arque-Bera 2.717216
Pr obab ility 0.2570 18
1
0
0
-20000
-10000
0
10000
-0.4
20000
Fig. 11 - Winning Bid/ Investment Regression
12
Series: Residuals
Sample 1 81
Observations 81
10
Me a n
- 2 .4 6 e - 1 7
Me d ia n
- 0 .0 0 3 8 8 6
Ma x imu m
0 .0 6 7 4 9 4
Min imu m
- 0 .0 5 9 1 0 7
Std . D ev .
0 .0 3 2 2 0 6
Sk e w n e s s 0 .1 8 6 7 6 7
Ku r to s is
2 .1 6 5 2 5 2
8
6
4
J a r q u e- Be r a 2 .8 22 6 1 9
Pr o b a b ility 0 .2 4 3 8 2 4
2
0
-0.06
Mean
Median
Max imum
Minimum
Std . D ev .
Sk ew nes s
Kurtos is
6
-0.04
-0.02
-0.00
0.02
0.04
0.06
-0.2
-0.0
0.2
0.4
Fig. 12 - Wining Bid Regression
14
Series: Residuals
Sample 1 81
Observations 81
12
10
Mean
Median
Max imum
Minimum
Std. D ev .
Sk ew nes s
Kurtos is
8
6
4
J arque-Bera 1.172691
Probability
0.556357
2
0
-20000
-3.67e-12
-339.2237
24098.23
-19168.80
8586.683
0.274548
3.214379
-10000
0
10000
20000
* We used in this section the EViews 5.0 package.
7.3
Principal Components Analysis
In this section, we ran a principal component analysis to deal with the multicollinearity between investment and
the reserve price in the Brazilian transmission auctions.
Principal components analysis (PCA) models the variance structure of a set of observed variables using linear
combinations of the variables. These linear combinations, or components, may be used in subsequent analysis, and
the combination coe¢ cients, or loadings, may be used in interpreting the components. While we generally require
as many components as variables to reproduce the original variance structure, we usually expect to account for
most of the original variability using a relatively small number of components.
PCA is mathematically de…ned as an orthogonal linear modi…cation that transforms the data to a new coordinate
system such that the greatest variance by any projection of the data comes to lie on the …rst coordinate (called the
…rst principal component), the second greatest variance on the second coordinate, and so on. PCA is theoretically
the optimum transform for given data in least square terms. Because of these transformations, PCA deals with the
multicollinearity problem.
The principal components of a set of variables are obtained by computing the eigenvalue decomposition of
the observed variance matrix. The …rst principal component is the unit-length linear combination of the original
variables with maximum variance. Subsequent principal components maximize variance among unit-length linear
combinations that are orthogonal to the previous components.
For details see Johnson and Wichtern (1992).
28
Table 7 - Cumulative Proportion of Principal Components Analysis
Computed using: Ordinary correlations
Extracting 11 of 11 possible components
Eigenvalues: (Sum = 11, Average = 1)
Number
Value
Difference
Proportion
Cumulative
Value
Cumulative
Proportion
1
2
3
4
5
6
7
8
9
10
11
4.971766
1.482532
1.369373
1.158753
0.889638
0.681831
0.265534
0.107199
0.056932
0.016441
9.61E-17
3.489233
0.113159
0.210619
0.269116
0.207807
0.416297
0.158334
0.050267
0.040491
0.016441
---
0.4520
0.1348
0.1245
0.1053
0.0809
0.0620
0.0241
0.0097
0.0052
0.0015
0.0000
4.971766
6.454298
7.823671
8.982425
9.872062
10.55389
10.81943
10.92663
10.98356
11.00000
11.00000
0.4520
0.5868
0.7112
0.8166
0.8975
0.9594
0.9836
0.9933
0.9985
1.0000
1.0000
The table 7 summarizes the eigenvalues, showing the values, the forward di¤erence in the eigenvalues, the
proportion of total variance explained, etc. Since we are performing principal components on a correlation matrix,
the sum of the scaled variances for the eleven variables is equal to 11. The …rst principal component accounts for
45.2% of the total variance (4.9718/11.00 = 0.4520), while the second accounts for 13.5% (1.4825/11.00 = 0.1348)
of the total. doind the same with the third and fourth components, we see that the …rst four components account
for over 81% of the total variation.
29
Table 8 - Linear Combination Coe¢ cients and Correlations
Eigenvectors (loadings):
Variable
PC 1
BID
0.431807
EPC
0.094776
INVEST
0.428454
EXT
0.406032
RAP_0
0.429030
REGFOUR
0.252089
REGONE
-0.153776
REGTWO
-0.075471
REGTHREE -0.009422
TAX
-0.400293
COMPET
-0.137272
Ordinary correlations:
BID
EPC
INVEST
EXT
RAP_0
REG4
REG1
REG2
REG3
TAX
COMPET
BID
1.000000
0.194756
0.910590
0.858417
0.935116
0.435932
-0.250908
-0.130147
-0.032704
-0.872141
-0.356776
PC 2
-0.010649
0.222400
-0.033723
0.073345
-0.024950
0.106279
0.424043
-0.779041
0.368316
0.072223
0.066330
EPC
1.000000
0.112302
0.151611
0.173027
-0.022119
-0.012056
-0.107418
0.164243
-0.170615
-0.271630
PC 3
-0.022650
-0.486674
0.063173
0.086098
-0.006446
0.379784
0.342498
-0.141749
-0.591765
0.038906
0.343303
INVEST
1.000000
0.863513
0.975626
0.450769
-0.274059
-0.111736
-0.045551
-0.853139
-0.164215
PC 4
-0.090791
-0.399158
0.042591
0.089422
-0.019523
0.159599
-0.516692
-0.085837
0.485785
0.043749
0.533412
EXT
1.000000
0.857330
0.532762
-0.260396
-0.227561
-0.007353
-0.698527
-0.135665
PC 5
PC 6
0.102375 -0.102668
-0.116554 0.727243
0.205159 0.004924
0.054752 0.095386
0.235097 -0.022110
-0.685764 0.158654
0.395282 -0.098687
0.152139 0.136068
0.113657 -0.223176
-0.154295 0.055876
0.434954 0.587035
RAP_0
REG4
1.000000
0.383555 1.000000
-0.252600 -0.295901
-0.099712 -0.388514
-0.012567 -0.265197
-0.855852 -0.380820
-0.232643 -0.082514
PC 7
0.036244
-0.026334
0.031436
0.637168
0.076973
-0.119453
0.009677
0.095566
0.000278
0.739654
-0.120471
PC 8
-0.218857
0.038775
0.529200
-0.523101
0.469060
0.088098
-0.034268
-0.046248
-0.000278
0.404652
-0.045559
PC 9
0.838758
0.006772
-0.239702
-0.348953
0.017238
0.039555
-0.034833
-0.004563
0.001218
0.300655
0.154021
PC 10
0.175854
0.025808
0.654884
-0.029458
-0.729993
-0.043809
0.014394
0.010946
0.017351
0.054477
-0.016283
REG1
REG2
REG3
TAX
1.000000
-0.402367
-0.274653
0.285309
0.101300
1.000000
-0.360616 1.000000
0.055876 0.027263 1.000000
-0.034335 0.019365 0.265272
PC 11
-2.30E-16
-5.03E-17
-1.45E-15
4.43E-16
1.20E-15
0.484243
0.493548
0.553954
0.463739
-8.78E-17
1.34E-17
COMPET
1.000000
Table 8 describes the linear combination coe¢ cients. We see that the …rst principal component (labeled "PC1")
is a roughly-equal linear combination of EXTENSION, INVESTMENT, RESERVE PRICE, TAX and BID variables;
The PC2, for its turn, is a combination of the dummies variables for South, Southeast and Northeast regions. The
PC3 is a combination of the EPC dummy, the dummies for the South, Northeast and Center West- North regions
and the variable of competitors number. We see that the regions are relevant for this group of variables, which
constitute our models to explain the transmission winning bids.
30
Figure 13 - Scree Plot and Cumulative Proportion
Scree Plot (Ordered Eigenvalues)
5
4
3
2
1
0
2
4
6
8
10
Eigenvalue Cumulative Proportion
1.0
0.8
0.6
0.4
0.2
0.0
2
4
6
8
10
The scree plot in the upper portion of the …gure 13 presents the sharp decline between the …rst and second
eigenvalues. Also depicted in the graph is a horizontal line with the mean value of the eigenvalues (which is always
1 for eigenvalue analysis conducted on correlation matrices, as we had done).
The lower portion of the graph shows the cumulative proportion of the total variance. As we saw in the table, the
…rst four components account for about 81% of the total variation. The diagonal reference line o¤ers an alternative
method of evaluating the size of the eigenvalues. The slope of the reference line may be compared with the slope
of the cumulative proportion; segments of the latter that are steeper than the reference line have eigenvalues that
exceed the mean.
7.4
Non-Linearity test
31
A possible di¤erence in the competitors behaviour can arise when we consider bids below or up R$ 48 millions. As
we already mentioned, companies with annual revenue below this value have an assumed pro…t income tax system
and, if they can, they will do the option to stay in this bene…cial tax system. Then we can expect a distinct
behaviour, more agressive for bids below this value, comparative to other bids. This is the hyphotesis we will test in
this non-linearity exercise. For this test, we choosed the …rst reduced-based model showed, once the TAX variable
appears more statistically signi…cant in this model.
Besides our sample being cross-section data, the test is based in Threshold Autoregressive (TAR) models, which,
according to Enders (2007) are a straighforward way to extend models to a non-linear format, with the possibility
to observe two regimes acting in the data.
According to results of this test in table 9, we can see that the exogenously determined limit, R$ 48 millions, is
relevant for the winning bid maginitude.
Table 9 - Regression with Exogenous Threshold
Dependent Variable:
2
Adjusted R :
Winning Bid
0.978
Residual Sums of Squares:
F test (16,42):
1439.82
119.12
Signi…cance level of F test:
0.0000
Durbin-Watson Statistics:
2.26
Table 10 - Estimation Results of Exogenous Threshold - R$ 48 millions as limit to Winning Bid
Variable
1. DEP
2. INVEST_DEP
3. RECEITA_DEP
4. EXT_DEP
5. EPC_DEP
6. COMP_DEP
7. SUL_DEP
8. SE_DEP
9. NE_DEP
10. CO_DEP
11. AP
12. INVEST_AP
13. RECEITA_AP
14. EXT_AP
15. EPC_AP
16. COMP_AP
17. SUL_AP
18. SE_AP
19. NE_AP
20. CO_AP
Coeff
2.15863608
-0.02857833
0.85265220
1.21964029
1.66403127
-1.14889582
0.20312681
-0.63975014
-0.76403728
0.00000000
31.96028507
0.05598604
0.15103367
5.05224044
0.00000000
-7.76259619
0.69804042
8.41913521
-4.90301495
0.00000000
Std Error
7.36863847
0.03640533
0.20610147
1.33801049
6.33795434
0.39975157
4.15545083
4.05168852
4.11475059
0.00000000
5.92386481
0.02327946
0.10176345
0.77498263
0.00000000
0.66181614
5.28644914
3.77329123
4.67004663
0.00000000
32
T-Stat
0.29295
-0.78500
4.13705
0.91153
0.26255
-2.87402
0.04888
-0.15790
-0.18568
0.00000
5.39517
2.40495
1.48416
6.51917
0.00000
-11.72923
0.13204
2.23124
-1.04989
0.00000
Signif
0.77100359
0.43685700
0.00016513
0.36721717
0.79418086
0.00633377
0.96124504
0.87529454
0.85358760
0.00000000
0.00000293
0.02065946
0.14523210
0.00000007
0.00000000
0.00000000
0.89558056
0.03105960
0.29977448
0.00000000
Notwithstanding, to verify if the better winning bid level that separate the agents behaviour is, in fact, R$
48 millions, we also tested the existence of a endogenous threshold. Table 11 and 12 show that this endogenous
threshold is statistically signi…cative and even more expressive than the exogenous limit (the comparison was made
with the Residual sums of Squares, that is lower in the second exercise). In this case, the estimated value for our
sample of the threshold was R$ 66.49 millions.
Table 11 - Regression with Endogenous Threshold
Dependent Variable:
Winning Bid
Adjusted R2 :
0.984
Residual Sums of Squares:
F test (16,42):
1058.15
163.03
Signi…cance level of F test:
0.0000
Durbin-Watson Statistics:
1.528
Table 12 - Estimation Results of Endogenous Threshold
Variable
1. DEP
2. INVEST_DEP
3. RECEITA_DEP
4. EXT_DEP
5. EPC_DEP
6. COMP_DEP
7. SUL_DEP
8. SE_DEP
9. NE_DEP
10. CO_DEP
11. AP
12. INVEST_AP
13. RECEITA_AP
14. EXT_AP
15. EPC_AP
16. COMP_AP
17. SUL_AP
18. SE_AP
19. NE_AP
20. CO_AP
Coeff
2.24793058
-0.04666323
0.85466818
1.18092056
3.83210549
-1.43262037
0.79221775
0.24099320
-0.38623135
0.00000000
44.20804816
0.05073948
-0.01982717
6.71997004
0.00000000
-7.36748303
-11.65150500
5.21958773
-2.53538717
0.00000000
Std Error
6.04106271
0.02895996
0.14430785
0.86814054
5.36791228
0.32577989
2.93061312
2.85863878
3.00286949
0.00000000
7.62859470
0.02125697
0.09852866
0.79200104
0.00000000
0.78442292
6.70619023
4.26121127
4.22522867
0.00000000
T-Stat
0.37211
-1.61130
5.92253
1.36029
0.71389
-4.39751
0.27032
0.08430
-0.12862
0.00000
5.79504
2.38696
-0.20123
8.48480
0.00000
-9.39223
-1.73743
1.22491
-0.60006
0.00000
Signif
0.71168308
0.11460470
0.00000051
0.18099562
0.47924313
0.00007327
0.78823426
0.93321571
0.89827168
0.00000000
0.00000078
0.02156798
0.84148819
0.00000000
0.00000000
0.00000000
0.08964110
0.22743585
0.55169087
0.00000000
Figure 14 shows the value that minimizes the residual sums of squares between all the possible thresholds. As
we already said, this value was R$ 66.49 millions.
33
Figure 14 - Residual Sums of Squares in Function of the Thresholds
* We used in this section the Rats 6.0 package.
8
Modelling an Economic Experiment to Capture the Synergy Gains
The econometric analysis with a reduced-based models approach, as presented before in this article, is a …rst attempt
to investigate the mechanism design applied in the Brazilian transmission lines auctions. It’s important to rebound
that structural econometrics is still in development, with models that do not include heterogeneous multiple-objects
auctions, like the transmission auction.
With this in mind, was necessary, as a way to complement this study, model an economic experiment to compare
the performance of this current auction design - a sequential auction of multiple objects - with another design, in
wich objects are o¤er in a simultaneous way, with some combination between them. This alternative design could
capture the revenue that competitors have in reason of the synergies between transmission lines contiguous or close
one another.
The winner is chosen in two phases. In the …rst phase, quali…ed bidders make sealed bids (an annual revenue to
construct and mantain the service) for the transmission lines. In this one, the bidder submitting the lowest value is
ranked as the best. In case there are bidders whose bid proposals were su¢ ciently close to the best ranked bidder’s
(in this case, higher by up to 10%), they shall be entitled to take part in the second phase as long as they accept
the revenue proposed by the best ranked bidder.
In the second phase, quali…ed bidders would compete in English auctions for the concession through annual
revenue proposals to obtain the concession granting. The winner would be the one who, upon accepting the lowest
revenue proposed in the …rst phase, makes a smaller bid in this phase. The Resolution that determines the bidding
mode states that in the second phase the granting proposals should be presented in successive bids.
A key point in establishing the bidding environment is that in some cases the segments to be auctioned are
adjacent or contiguous; i.e. connected segments shall be assigned. Furthermore, some of these concessions refer to
segments connected to others operated by utilities whose shareholders are companies that may take part in current
34
biddings. This means that such environment is made up by both synergies among the objects to bid or between
these and others whose operation has already been assigned in the past. In this case, there is asymmetry among
bidders.
The concession bidding of transmission lines segments is an example of multiple heterogeneous objects auctions.
A key element to increase the complexity in this environment is the fact that more than one object can be assigned
to each bidder (in this case, the concession agreement). Furthermore, the environment is asymmetric since there
are players operating licenses connected to others that shall be bid. In the presence of synergies, the winner of a
component making up a certain basket of goods obtains advantages vis-à-vis the others. Such asymmetry may lead
to revenue loss for the auctioneer if objects are sequentially auctioned. In this case, there is an income transfer
from society, in general terms, to a private player thus impairing the reputation of the policy maker20 .
In the existence of complementarities among objects, i.e. if the items are not independent, the task of bids
submission becomes di¢ cult as a result of concession assignment by means of a sequence of auctions. Therefore, it
is worth investigating auction formats that allow players to submit bids that not only embody such synergies, but
also attract bidders.
It is a fact known in the auction literature21 that the auction format a¤ects the result both in terms of e¢ ciency
(capacity of assigning the object to the player who best values it) and revenue collection or cost reduction (in case of
descending auctions). Therefore, the environment con…guration allows recommendations regarding auction formats
that best …t a speci…c context. However, such recommendations might have a greater implementation probability
if they could be illustrated through situations able to reproduce the incentives existing in the environment under
analysis.
This study reports the experimental analysis of alternative auction mechanisms, which means the electronic
implementation of alternative formats of multiple objects auctions. The group of implemented auctions is the result
of theoretical studies as well as the assessment of auction experience.
8.1
The Model
The experiment carries out the comparative investigation of two mechanisms subject to use in the allocation of
multiple heterogeneous objects: a hybrid auction of a single object used sequentially implemented and a simultaneous
auction of multiple rounds. The performance of the said mechanisms is described below.
In each session, a group of I players, i = 1; ::::; I, competes for a group of K heterogeneous goods. The objects
are auctioned either sequentially or simultaneously. As an example of the Brazilian transmission lines concessions,
the number of objects to be auctioned equals eight. For each license taken individually, each bidder attributes a
private value independently distributed as follows:
vi (lk )
U [v k ; v k ] :
This piece of information is common knowledge at the beginning of the experimental session.
2 0 Paul Kemplerer reports examples of ‡aws in auction design harming the reputation of economic policy makers. For further
references, see Paul Kemplerer, “Auction: Theory and Practice”, Princeton University Press, 2003.
2 1 For further references, see Paul Milgrom, “Putting Auction Theory to Work”, Cambridge University Press, 2003 and Flavio M.
Menezes and Paulo K. Monteiro, “An Introduction to Auction Theory”, Oxford University Press, 2004.
35
The design envisages the existence of scope economies or synergies between subsets of objects. The existence of
positive synergies for a given subset of commodities means that to some pre-de…ned baskets the value of the joint
possession of objects is higher than the sum of values separately assigned to the objects. Let S K be a subset of
licenses presenting positive synergies. Without loss of generality, assume that for the entire subset s 2 S;
vi
[ lk
k2s
=
s
X
!
vi (lk )
k2s
k
X
j=1
where vi (l1 ; ::; lj ) is bidder i s value for the basket (l1 ; ::; lj ) and
8.1.1
s
vi (lj ) 8lk 2 s
(10)
> 1:
Treatments
The study tests two alternative mechanisms: a sequential auction and a simultaneous ascending auction.
Sequential Auction (Sequence of hybrid auctions for a single object) In this treatment, the objects are
sequentially auctioned through an auction format similar to the one used in the Brazilian privatization process22 .
This is a sequential auction where one object is assigned in each auction. One can think of the object either as a
license for the concession of a transmission line or as the right to explore a power generation development.
All players (potential buyers) submit sealed bids for a given object. After the opening of bids, if there are players
whose bids are su¢ ciently close to the highest (lowest) bid, the ones quali…ed to take part in the second phase will
be the bidder with the highest (or lowest) bid and those bidders whose bids are higher (or lower) than or equal to
times the value of the highest (lowest) bid. In the experiment = 0; 9. Thus, if there are bids that di¤er from the
winning bid by less than 10%, the quali…ed bidders compete for the object in an ascending auction. In this auction,
the reserve price, or the lowest (higuest) admissible bid, is the highest (lowest) bid submitted in the …rst phase. In
order to take part in the second phase, quali…ed bidders should be willing to honor the payment represented by the
highest (lowest) bid of the …rst phase.
The second phase consists of an ascending auction23 . In this auction, at each period the current price is equal
to the current price of the preceding period plus (or minus) a minimum increment denoted by min . Bidders may
accept or reject the current price; if they reject it, they will be out of the bidding. The default situation is to accept
the bid. The auction ends with one remaining player who is then declared the winner. The "price" to be paid is
the price to which the next-to-last player leaves the auction.
Simultaneous Ascending Auction Besides the sequential auction, in a second treament the K licenses are
allocated through a simultaneous ascendind auction.
Simultaneous auctions have been championed and extensively used to assign commodities in cases where the
values assigned by bidders are not independent. Although they might be susceptible to exposure problems, the
relative simplicity of the bids formulation process makes this auction format proper for scenarios where it is desirable
that players have an opportunity to reach a desired object aggregation.
2 2 For
2 3 For
further references, see Dutra, J. and F.M. Menezes, "Hybrid Auctions", Economics Letters 77, 301-307, 2002.
references on auctions, see F.M. Menezes and P.K. Monteiro “An Introduction to Auction Theory”, Oxford University Press,
2004.
36
In the auction24 , the bids by players at each period must meet the budget constraint, i.e. the sum of their bids
K
X
for the set of commodities they are competing for,
bti(lk ) ; cannot exceed the initially assigned income. Denote
k=1
by w0i s the initial endowment of the i-th player. Its budget constraint is given by:
K
X
bti(lk )
k=1
w0i ; i = 1; :::; I; 8t
(11)
In order to avoid demand reduction, it is common practice to adopt an activity rule. In this case, the rule is
a monotonicity requirement in the number of objects for which the bidder competes. Let Nit be the number of
elements in the set of objects for which the i th player submitted a bid in period t. Then the monotonicity rule
requires that
Nit
Nit
1
8t
(12)
an i player who is said to be active for a set of licenses cannot increase the number of licenses to which the bid is
submitted. Furthermore, at each new period only the bids meeting the following condition shall be accepted
bti;k
n
o
max bti;k1 +
i
min
k ;
k = 1; :::; K; i = 1; :::; I
(13)
i.e. price bids (in this case, granting value) must be non-decreasing. However, the bidder submitting the highest
(lowest) bid in a t round does not have to propose a higher (lower) bid in the subsequent round. In this case, at
the beginning of the (t + 1) th round this one should be considered as active.
Conditions (11) (13) must be met at each round so that player i may continue taking part in the auction.
The end of the auction is simultaneous; i.e. the auction continues as long as bids are still being submitted by at
least one player for at least one object. That means the auction goes on while there is a price change for at least
one product. For each object, the winner would be the bidder who would have submitted the highest (lowest) bid
at the moment the auction is over; prices to be paid will be the ones in force.
8.1.2
Auction Rules
Before the beginning of the auction, each bidder is informed of:
the values assigned to each commodity;
the values associated with pre-established combinations for which there are positive synergies (transmission
lines close one to another).
Private Values The …rst set of auctions is characterized by private values: for each object bidder i0 s value is a
number extracted with equal probability of a distribution that is of common knowledge.
In the case of the experimental design, it was established the existence of synergies for licenses 3, 4 and 5 as
well as for the 7 and 8 combination; the following valuations - just a possibility of gains - for commodities were
2 4 This
restriction applies to the sequential auction as well.
37
experimentally implemented,25
vi
vi
5
[ lk
k=3
8
[ lk
k=7
= 1:44
5
X
vi (lk )
k=3
= 1:2
8
X
vi (lk )
(14)
k=7
In 11 auctions the values assigned to the objects by bidders were private ones. In this case, all values were
extracted with equal probability from the [30, 80] interval and such fact was common knowledge.
Almost-Common Values In this second set of experimental sessions, a di¤erent form of assigning value to
commodities was implemented. This structure describes the case where the value of the object has a common
component for all bidders and this value is not known at the moment bids are submitted. The auctions motivating
the current analysis are part of this group; i.e. there is a common component associated with the value of the object
that is being sold.
For sessions numbered 12 to 17, players’values for the objects by were made up of two parts. In each auction,
the value that participant i assigns to the k th commodity, Vi;k , is equal to the sum of a common value component
(ck ), which is equal to all bidders, and of a private value (xi;k ), as follows:
Vi;k = xi;k + ck
The value of the common component of each commodity, ck , k = 1; :::; 8, is a number extracted with equal
probability from the interval [ck ; ck ]; however, such value is not known until the end of the auction. The bidder
observes only a signal, sk , which can be understood as an estimate of the value of this common component; for each
commodity, this signal si;k is uniformly distributed in the interval [ck z; ck + z] and this is common knowledge
The private signal xi;k is a number uniformly distributed in the [xk ; xk ] interval. Note that positive synergies for
referred sets are still valid.
TESTABLE HYPOTHESIS: In the presence of positive synergies, the simultaneous auction allows bidders with
a higher probability to bene…t from such synergies, thus assuring higher e¢ ciency at the auction and higher revenue
for the auctioneer.
8.1.3
Payo¤s
Besides observing the information concerning values for the goods, at the beginning of the auction each bidder
observes the initial prices of each object. Upon such information, the auction gets started.
In each experimental session, the players’ earnings are made up of a participation fee, in the form of a …xed
rate, plus their decision gains throughout the session. For each acquired commodity, the winner’s gain is given by:
vi (lk )
pk = vi (lk )
max bti;k
i2I
extension of the study leads to the analysis of the case of a group of players who hold a D license previously assigned to a bi
participant in the auction so that vbi (C + D) > vbi (C) + vbi (D) : This example characterizes the presence of asymmetries among players.
2 5 An
38
where bti;k = bti (lk ). The non-winning bidders earn nothing. Total gains in the session are equal to gains for the
sum of obtained licenses, net of paid prices.
If the bidder managed to add the corresponding licenses to pre-established synergies, his gain for said combination
is equal to the value for the combination, which is given by (10), net of the price paid for the combination, which is
equal to the sum of paid prices. By way of illustration, if participant 1 won licenses 3, 4 and 5 his gain, 1 , would
be equal to:
1
(l3 ; l4 ; l5 )
= v1 (l3 ; l4 ; l5 )
=
(p3 + p4 + p5 )
1:44 (v3;1 + v4;1 + v5;1 )
(p3 + p4 + p5 )
At the end of the experimental session, the payment is made in cash.
8.2
8.2.1
Experiments Results
Aggregated Results
Seventeen experimental sessions have been carried out. Out of these, 8 sessions consist of sequential auctions while
the remaining ones are simultaneous auctions. In each auction, a group of bidders ranging from six to eight competes
for the ownership of eight commodities. As a whole, 136 commodities were auctioned, which stood for licenses.
The values assigned by bidders to the objects were private ones in 88 auctions. In this case, all values were
extracted with equal probability from the [30; 80] interval and such fact was of common knowledge.
For auctions numbered 89 to 136, in turn, the values assigned by players to the objects were made up of two
parts (almost-common values); in each auction, the value that bidder i assigns to the k th commodity, Vi;k , is
equal to the sum of a common value component (ck ) and a private value (xi;k ) as follows:
Vi;k = xi;k + ck
The common value component of each commodity,ck , i = 1; :::; 8, is a number between 30 and 80, drawn with
equal probability; however, this value is not observed by i: The bidder knows only a signal, si;k , which can be
understood as a value estimate of this common component. For each commodity, this signal, si;k ; is uniformly
distributed in the [ck 10; ck + 10] interval. On the other hand, the private signal xi;k is a uniformly distributed
number in the [0; 30] interval.
Table 13 summarizes the results of the experimental design. The following conclusions can be drawn from said
data:
The mean e¢ ciency of the simultaneous auction is relatively higher than that of the sequential auction;
The revenue obtained with the implementation of the simultaneous auction is higher;
The number of ine¢ cient assignments (number of times in which the winner of the auction was a player who
would not present the highest (lowest) value for the commodity) is lower in the simultaneous auction;
39
Mean gains of bidders in the simultaneous auction are comparatively higher than those in the sequential
auction.
Table 13 - Experimental Results (Mean per Treatment)
Pro…t/Potential Value1
Mechanism
Values
Sequential
Private
Sequential
Almost Common
Simultaneous
Private
Simultaneous
Almost Common
Value Appropriation2
4%
8%
3%
2%
92%
97%
97%
99%
Notes: (1) Bidders’mean pro…t as proportion of the maximum possible value;
(2) Value realized as proportion of the maximum possible value.
According to results, the simultaneous auction is higher both in terms of e¢ ciency and revenue guarantee
(without necessarily resulting in losses to bidders).
This is the conclusion of the experimental study carried out in an environment meant to reproduce characteristics
that possible exists in transmission lines scenario auctions.
8.2.2
Individual Behavior
Analysis results of bidders’individual behavior are presented below. Table 14 shows data on the bids as proportion
of players’values in the auction. It is inferred that players are more aggressive in the case of simultaneous auctions
and such behavior is more intense the higher the synergy degree among goods. As described in the experimental
design, the absence of synergy refers to goods 1, 2 and 6; while the weak and strong synergies refer to baskets (l7 ; l8 )
and (l3 ; l4 ; l5 ) respectively.
Table 14 - Bid Behavior (Bid as value proportion).
Mechanism
Sinergy
Absent
Weak
High
Private Values
Almost-Common Values
General Mean
Sequential
Simultaneous
Sequential
Simultaneous
Sequential
Simultaneous
0:93
0:88
1:02
0:84
0:97
1:21
0:91
0:95
1:00
0:74
1:03
1:23
0:93
0:94
1:03
0:81
0:99
1:22
A more aggressive bidding behavior is favored by the auctioneer’s once it allows a higher appropriation of the
value bidders assign to the objects. In the case of transmission line concessions this means a higher rent extraction
in the form of a decrease in the revenue they accept to construct and administrate the transmission lines. Table
15 shows the percentage variation of bids as value proportion relatively to the case in which goods do not present
synergy. Data denote an advantage of the simultaneous auction over the sequential one; an advantage that is even
more stressed in the case of almost-common values.
40
Table 15 - Bid as value proportion (% Variation).
Mechanism
Sinergy
Weak
High
Private Values
Sequential
Almost-Common Values
General Mean
Simultaneous
Sequential
Simultaneous
Sequential
Simultaneous
13:00
44:05
4:00
9:89
29:00
66:22
1:00
10:75
18:00
50:62
5:00
9:68
Base: Absence of synergy
Mean equality tests for the case of private values are shown in Table 16. At a 5% level, means are statistically
di¤erent for goods 1 and 2 (absence of synergy) and for good 4.
Table 16 - Means Di¤erence Test – Private Values
Mean per Mechanism
Good
Good 1
Good 2
Good 3
Good 4
Good 5
Good 6
Good 7
Good 8
Sequential
Simultaneous
p- value
0:978
0:980
1:194
1:009
0:954
0:871
0:990
0:875
0:836
0:841
1:251
1:279
1:113
0:833
0:992
0:955
0:03
0:00
0:61
0:02
0:16
0:65
0:98
0:34
In the case of almost-common values, data reported in Table 17 show an advantage of the sequential auction in
the case of no synergy (Goods 1 and 6). On its turn, at the 5% signi…cance level, the simultaneous auction reveals
a more aggressive bidding behavior under strong synergy. Note that in the case of Good 3, said means do not
di¤er under the statistical viewpoint. Such behavior is expected considering that in the sequential auction players
show more aggressiveness in the auction of the …rst object of the basket. This strategy ends up by reducing the
participation of other players in subsequent auctions (of basket objects) since only the winner of commodity 3 may
then carry out the involved synergies. Figure 2 of Appendix A2 shows these arguments.
Table 17 - Means Di¤erence Test – Almost-Common Values
Mean per Mechanism
Good
Good 1
Good 2
Good 3
Good 4
Good 5
Good 6
Good 7
Good 8
Sequential
Simultaneous
p- value
0:894
0:878
1:109
0:949
0:950
0:935
0:996
0:947
0:658
0:799
1:202
1:232
1:262
0:734
1:018
1:087
0:00
0:33
0:56
0:02
0:01
0:00
0:77
0:19
41
The analysis of individual data reveals a comparative edge of the simultaneous auction format vis-à-vis the
sequential one. In the presence of positive synergies, bids by players in the simultaneous auction re‡ect the increase
of the value associated with the ownership of pre-established combinations of commodities. In the sequential auction,
in turn, the uncertainty as to the capacity of reaching the desired aggregation of commodities compel players to
adopt a more conservative behavior in which joint ownership is not priced. Figures 1 and 2 of appendix A2 show
these observations.
Finally, a regression analysis of bids is shown in Tables A5, A6 and A7. The inclusion of several interaction
dummies between the simultaneous mechanism and goods con…rm the presented arguments especially in the case
of almost-common values (Table A6 attached). Such evidence takes on greater importance taking into account that
this value structure is closer to reality.
9
Conclusions
Our empirical exercises suggest that many factors a¤ect, in the direction of economic intuition, the winning bid in
transmission auctions. We modeled four reduced-based models and found that a better speci…cation is to use the
winning bid in level against the logarithm and a ratio of the bid with investments need.
With these models, it seems that the number of competitors in the auctions, in all reduced-based models, a¤ected
in a negative way the level of the winning bid; the geographic location of the line is also important to the winning
bid, so our hypothesis that a company operating a line in a determined region will give more agressive bids in
auctions for lines close to the …rst lines of this company becomes relevant (see the Chesf case); the dummy variable
indicating an o¤er below R$ 48 millions is signi…cative in the estimation and, as a consequence, we can say that
…rms concede a pro…t margin to have a di¤erentiated tax system; the projects characteristics, as such the extension
of the lines and the estimated investment to the construction and maintanance also contribute to the winning bid
level.
The reduced-based model that applied the model from Rezende (2005) let us to estimate the dictribution function
of the competitors’valuation. The results indicate a logistic distribution to the competitors valuation. The logistic
distribution has a shape very similar to that of the Normal distribution, with the mean as one of its parameters.
Since the logistic distribution is symmetrical, the median and the mode are always equal to the mean.
In our Principal Component Analysis, we found that the main variables responsible for the winning bid are the
extension, the investment, the reserve price of the auction, the dummy relative do the tax system. However, we also
see that the regions are relevant for this group of variables, which constitute our models to explain the transmission
winning bids.
For its turn, the non-linearity analysis, which was applied with the objective to test if our model has two regimes,
identi…es that both exogenous (R$ 48 millions) and endogenous thresholds for the winning bids are valid to our
analysis. As a consequence, we can say that competitors have a di¤erent behaviour in each subsample of our data.
The endogenous threshold was the best especi…cation for this non-linear analysis, indicating that bids are di¤erent
beyond and up R$ 66 millions.
Finally, we model an economic experiment testing that an alternative design for Brazilian transmission auctions,
with neighbor lines auctioned simultaneously would improve the rent extraction from the transmission companies.
42
The current mechanism for the bidding is a sequential auction: the concessions would be allocated in a predetermined order. However, two elements provide uniqueness to the concession environment. Firstly, the bidding
involves some segments that are connected or close to others that have already been bid. Estimates by the Ministry
of Finance report the existence of considerable scale economies in the joint operation of connected segments.
Furthermore, some involved segments are close to others that were the object of concessions in the past. In
this case, there is asymmetry among bidders as such companies already holding concession agreements can operate
the segments under more favorable conditions (at lower cost).
Both elements presented herein justify the comparison of the auction format proposed by the Ministry of Transportation with a simultaneous auction. This comparison was carried out in an experimental environment developed
to reproduce conditions similar to the ones presented. The results con…rm a superiority of the simultaneous auction
whenever there are positive synergies: it was possible to simplify the bid submission task, thus obtaining higher
revenue (in the granting phase) and higher e¢ ciency (choosing the bidder who assigns a higher value to the object).
REFERENCIAS DA JOISA!!!!
References
[1] Athey & Haile (2001) “Identi…cation of Standard Auction Models,” MIT Working Paper 00-18.
[2] Athias & Nunes (2007), "Winner’s Curse in Toll Road Concessions", Economics Letters, forthcoming.
[3] Ausubel et al. (1997), “An E¢ cient Ascending-Bid Auction for Multiple Objects,” Working Paper No. 97-06,
University of Maryland.
[4] Bajari (1997), “The First-Price Auction with Asymmetric Bidders: Theory and Applications,” Ph.D. Dissertation (University of Minnesota).
[5] Bajari, P. (2000), "Econometrics of Sealed Bid Auctions", working paper, Harvard University.
[6] Bajari, Hortaçsu (2005), "Are Structural Estimates of Auction Models Reasonable? Evidence from Experimental Data", Journal of Political Economy, 113, pgs 703-741.
[7] Bajari & Ye (2001), "Competition Versus Collusion in Procurement Auctions: Identi…cation and Testing",
Working Paper Stanford University.
[8] Boucher et al (1999),
[9] Brandão, De Castro, (2006), "Os leilões de linhas de transmissão e o Risco Brasil", working paper, UFRJ.
[10] Bueno & De Castro (2006a) "Leilões de Linhas de Transmissão e o Modelo de Parceria Estratégica PúblicaPrivada", working paper, UFRJ.
[11] Bueno & De Castro (2006b) Análise e Pespectiva do Leilão de Linhas de Transmissão de Energia Elétrica de
novembro de 2006, working paper, GESEL/UFRJ.
[12] De Silva (2005) "Auctions: An Empirical Investigation", Economic Inquiry, Vol 73, no 1.
43
[13] Dutra & Menezes (2003), "Hybrid Auctions", Economics Letters 77, 301-397.
[14] Enders, Walter "A Threshold Model of Real U.S. GDP and the Problem of Constructing Con…dence Intervals
in TAR Models", working paper.
[15] Feinstein et al. (1985), “Asymmetric Information and Collusive Behavior in Auction Markets,” American
Economic Review, 75, 441-460.
[16] França, F, Ramos, F. (1998), "Decisão de Investimento na Expansão do Sistema de Transmissão Face a Reestruturação do Setor Elétrico", ENEGEP.
[17] Francellino & Polito (2007), O segredo ‡amenco, Brasil Energia, 2007.
[18] Gandal (1997), "Sequential Auctions of Interdependent Objects: Israeli Cable Television Licenses", Journal of
Industrial Economics, 1997.
[19] Gilley and Karels (1981), “The Competitive E¤ect in Bonus Bidding: New Evidence,” Bell Journal of Economics: 12:637 - 648.
[20] Hair (2005)
[21] Harris & Raviv (1981), "Allocation Mechanisms and the Design of Auctions", Econometrica, Vol. 49, No. 6
(Nov., 1981), pp. 1477-1499.
[22] Hendricks & Porter (2003), “Empirical Implications of Equilibrium Bidding in First-Price, Symmetric, Common
Value Auctions,” Review of Economic Studies 70:115-145.
[23] Hirota (2006), O Mercado de Concessão de Transmissão de Energia Elétrica no Brasil, dissertação USP.
[24] Jofre-Bonet, M. and M. Pesendorfer (2000), “Bidding Behavior in Repeated Procurement Auctions,”European
Economic Review 44:1006-1020.
[25] Jofre-Bonet, Pesendorfer (2003), Estimation of a Dynamic Auction Game, Econometrica, vol 71 no 5.
[26] Johnson and Wichtern (1992).
[27] Klemperer, P. (2002), “What Really Matters in Auction Design”, Journal of Economic Perspectives, American
Economic Association, vol. 16(1), pages 169-189.
[28] Paarsch, H. (1991) “Empirical Models of Auctions and an Application to British Columbian Timber Sales,”
Research Report 9212 (University of Western Ontario).
[29] Porter, R. and J. D. Zona (1993), “Detection of Bid Rigging in Procurement Auctions,” Journal of Political
Economy 101: 518-538.
[30] Rezende & Garcia (2000), Leilões de Títulos da Dívida Pública pelo Banco Central do Brasil: Um Estudo dos
Fatores Condicionantes da Dispersão das Propostas para os BBCs , working paper, PUC-Rio.
[31] Rezende, Leonardo (2005), Auction Econometrics By Least Squares, Journal of Applied Econometrics, Volume
23 Issue 7, Pages 925 - 948.
44
[32] Eduardo Serrato M. Ribeiro, ANEEL.
[33] Rusco & Walls (1999) ,Competition in a repeated spatial auction market with an application to timber sales,
Journal of Regional Science 39 pp. 449–465
[34] Shum & Hong (2000), Econometric models of asymmetric ascending auctions, Journal of Econometrics 112.
[35] Wilson (1979), “Auctions of Shares,” Quarterly Journal of Economics 93:675 - 689.
[36] Wolfram (1998), “Strategic Bidding in a Multi-Unit Auction, An Empirical Analysis of Bids to Supply Electricity in England and Wales,” RAND Journal of Economics 29: 703-725.
10
Appendix - Econometyric Analysis - A1
Outcomes from estimated regressions trying to model the winning bids in the actual transmission auctions are
present below.
Table A1- Winning Bid as Dependent Variable
Dependent Variable: BID
Method: Least Squares
Date: 05/13/09 Time: 18:10
Sample: 1 81
Included observations: 81
White Heteroskedasticity-Consistent Standard Errors & Covariance
Variable
Coefficient
Std. Error
t-Statistic
Prob.
COMPET
REGONE
REGTWO
REGTHREE
REGFOUR
TRIBUTACAO
EXTENSAO
INVEST
-1585.067
38642.85
36498.68
36226.94
34407.36
-25889.79
41.21977
0.045189
311.6720
7529.648
7405.643
7285.152
7233.964
6739.321
14.28744
0.016769
-5.085690
5.132093
4.928495
4.972709
4.756363
-3.841602
2.885034
2.694856
0.0000
0.0000
0.0000
0.0000
0.0000
0.0003
0.0051
0.0087
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
0.920533
0.912913
9700.937
6.87E+09
-854.3006
1.665407
Mean dependent var
S.D. dependent var
Akaike info criterion
Schwarz criterion
Hannan-Quinn criter.
29236.41
32872.73
21.29137
21.52786
21.38626
Table A2 - Logarithm of Winning Bid as Dependent Variable
45
Dependent Variable: LBID
Method: Least Squares
Date: 05/11/09 Time: 16:30
Sample: 1 81
Included observations: 81
White Heteroskedasticity-Consistent Standard Errors & Covariance
Variable
Coefficient
Std. Error
t-Statistic
Prob.
LINVEST
LEXTENSAO
TRIBUTACAO
LCOMPET
0.857602
0.064119
-0.172994
-0.390184
0.013465
0.025504
0.057238
0.041543
63.69129
2.514049
-3.022373
-9.392280
0.0000
0.0140
0.0034
0.0000
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
0.950451
0.948520
0.261861
5.279987
-4.347740
0.969482
Mean dependent var
S.D. dependent var
Akaike info criterion
Schwarz criterion
Hannan-Quinn criter.
9.678662
1.154125
0.206117
0.324361
0.253558
Table A3 - Winning Bid/Estimated Investiment as Dependent Variable
Dependent Variable: BIDPERINVEST
Method: Least Squares
Date: 05/14/09 Time: 16:51
Sample: 1 81
Included observations: 81
Variable
Coefficient
Std. Error
t-Statistic
Prob.
COMPET
REGONE
REGTWO
REGTHREE
REGFOUR
-0.010686
0.209325
0.193918
0.185277
0.194548
0.001051
0.010297
0.008742
0.010473
0.009664
-10.16858
20.32806
22.18252
17.69171
20.13093
0.0000
0.0000
0.0000
0.0000
0.0000
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
0.583824
0.561920
0.033044
0.082984
163.8501
1.133266
46
Mean dependent var
S.D. dependent var
Akaike info criterion
Schwarz criterion
Hannan-Quinn criter.
0.131981
0.049924
-3.922225
-3.774419
-3.862923
Table A4 - Winning Bid as Dependent Variable
Dependent Variable: BID
Method: Least Squares
Date: 05/13/09 Time: 18:10
Sample: 1 81
Included observations: 81
White Heteroskedasticity-Consistent Standard Errors & Covariance
11
Variable
Coefficient
Std. Error
t-Statistic
Prob.
COMP1
COMP2
COMP3
COMP4
COMP5
COMP6
COMP7
COMP8
COMP9
COMP10
COMP11
COMP12
COMP13
COMP14
EXTENSAO
TRIBUTACAO
EPC
INVEST
30607.07
29923.89
25147.73
25421.69
20626.79
24048.61
16223.32
11336.19
12603.15
17090.75
17285.73
331.7599
19349.76
16282.07
41.01090
-22779.68
2405.861
0.051510
9460.806
9725.857
8979.119
8806.802
11379.26
9688.813
9701.743
9963.571
10516.03
9732.952
8494.027
12668.89
8488.785
8791.562
14.37188
7456.712
3979.275
0.018504
3.235144
3.076736
2.800690
2.886597
1.812665
2.482101
1.672207
1.137763
1.198470
1.755968
2.035045
0.026187
2.279449
1.852011
2.853552
-3.054922
0.604598
2.783701
0.0019
0.0031
0.0068
0.0053
0.0746
0.0157
0.0994
0.2595
0.2352
0.0840
0.0461
0.9792
0.0260
0.0687
0.0058
0.0033
0.5476
0.0071
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
0.931769
0.913358
9676.096
5.90E+09
-848.1263
1.819940
Mean dependent var
S.D. dependent var
Akaike info criterion
Schwarz criterion
Hannan-Quinn criter.
29236.41
32872.73
21.38583
21.91793
21.59932
Apppendix - Experiments - A2
Outcomes from estimated regressions aiming to assess the determinants of players’bids in the auction are presented
below.
47
Table A5 - Bid Determinants –Private Values
Variables
Private Value
Mechanism
Good 2
Good 3
Good 4
Good 5
Good 6
Good 7
Good 8
Mechanism *Good
Mechanism *Good
Mechanism *Good
Mechanism *Good
Mechanism *Good
Mechanism *Good
Mechanism *Good
Constant
2
3
4
5
6
7
8
Number of Observations
F(16; 607)
Prob > F
R2
Coe¢ cient
p- value
0:96
5:72
0:10
12:73
3:07
0:87
4:64
1:83
4:05
2:17
5:32
14:70
13:84
1:51
3:96
9:03
1:10
0:00
0:03
0:93
0:00
0:39
0:87
0:03
0:64
0:13
0:49
0:35
0:01
0:04
0:73
0:47
0:04
0:73
624
32:34
0:00
0:41
Notes: (1) Mechanism is a dummy with value 1 if the auction is simultaneous;
(2) bemi is a dummy variable with value 1 for the i th good, i = 2; :::; 8;
(3) M echanism
goodi is the interaction of the i-th good with the mechanism.
:
48
Table A6 - Bid Determinants –Almost-Common Values
Variables
Private Value
Mechanism
Good 2
Good 3
Good 4
Good 5
Good 6
Good 7
Good 8
Mechanism *Good
Mechanism *Good
Mechanism *Good
Mechanism *Good
Mechanism *Good
Mechanism *Good
Mechanism *Good
Constant
2
3
4
5
6
7
8
Number of Observations
F(16; 271)
Prob > F
R2
Coe¢ cient
p- value
1:30
24:62
0:37
18:92
7:04
7:92
8:87
6:05
3:66
6:19
18:51
30:76
30:32
2:74
14:98
18:11
49:96
0:00
0:00
0:92
0:00
0:14
0:06
0:03
0:31
0:50
0:40
0:05
0:00
0:00
0:65
0:06
0:02
0:00
288
17
0:00
0:44
Notes: (1) Mechanism is a dummy with value 1 if the auction is simultaneous;
(2) bemi is a dummy variable with value 1 for the i th good, i = 2; :::; 8;
(3) M echanism
goodi is the interaction of the i-th good with the mechanism.
49
Tabela A7. Bidding Determinants
Variables
Private Value
Mechanism
Good 2
Good 3
Good 4
Good 5
Good 6
Good 7
Good 8
Mechanism *Good
Mechanism *Good
Mechanism *Good
Mechanism *Good
Mechanism *Good
Mechanism *Good
Mechanism *Good
Constant
2
3
4
5
6
7
8
Number of Observations
F(16; 895)
Prob > F
R2
Coe¢ cient
p- value
0:05
15:74
1:77
12:72
5:90
1:86
2:21
0:50
5:83
5:03
12:89
18:43
26:87
3:42
12:16
18:18
58:60
0:10
0:00
0:58
0:01
0:15
0:70
0:55
0:91
0:13
0:31
0:04
0:00
0:00
0:53
0:04
0:00
0:00
912
7:97
0:00
0:12
Notes: (1) Mechanism is a dummy with value 1 if the auction is simultaneous;
(2) bemi is a dummy variable with value 1 for the i th good, i = 2; :::; 8;
(3) M echanism
goodi is the interaction of the i-th good with the mechanism.
The results of Table A.3 show that, in the presence of synergies the mechanism-good interaction presents a
signi…cant positive e¤ect from a statistical viewpoint. Asterisks indicate statistically signi…cant parameters at 5%
level.
Figures 1 and 2 from this appendix present bid data as a proportion of the value the bidders assign to the
objects in cases of private and almost-common values respectively.
50
Figure 1. Bid Behavior
Private Values
1,4
1,2
1
0,8
0,6
0,4
0,2
0
No Synergy
Weak
Strong
Sequential
0,93
0,88
1,02
Simultaneous
0,84
0,97
1,21
Figure 2 - Bid Behavior
Almost-Common Values
1,4
1,2
1
0,8
0,6
0,4
0,2
0
No Synergy
Weak
Strong
Sequential
0,91
0,95
1,00
Simultaneous
0,74
1,03
1,23
51