WPS4105

                  DETECTING COLLUSION IN TIMBER AUCTIONS:
                                AN APPLICATION TO ROMANIA


                JEAN-DANIEL SAPHORES, JEFFREY R. VINCENT, AND VALY MAROCHKO

                   WITH IOAN ABRUDAN, LAURA BOURIAUD, AND CLIFFORD ZINNES




Correspondence

        Saphores:         School of Social Ecology
                          University of California, Irvine
                          Irvine, CA 92697-7075
                          saphores@uci.edu

        Vincent:          Graduate School of International Relations & Pacific Studies
                          University of California, San Diego
                          9500 Gilman Drive
                          La Jolla, CA 92093-0519
                          jvincent@ucsd.edu


World Bank Policy Research Working Paper 4105, December 2006

The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange
of ideas about development issues. An objective of the series is to get the findings out quickly, even if the
presentations are less than fully polished. The papers carry the names of the authors and should be cited
accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors.
They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they
represent. Policy Research Working Papers are available online at http://econ.worldbank.org.


Acknowledgments

We thank, in alphabetical order, Peter Dewees for guidance and comments on various drafts,
Michael Kanaley for administrative support, Nalin Kishor for comments on the draft final report,
Liviu Nichiforel for data collection in Suceava, Leonard Padureanu for a background report that
provided much of the information in section 3, Robert Subrick for constructing the auction
datasets for Neamt and Suceava, and Clare Wolfowitz for editorial assistance. We also thank the
individuals and organizations in Romania listed in the appendix for their cooperation, the World
Bank and the U.N. Food and Agriculture Organization for financial support, and the University
of California's Institute on Global Conflict and Cooperation for administrative support.




                                                           1

                     DETECTING COLLUSION IN TIMBER AUCTIONS:
                               AN APPLICATION TO ROMANIA


Abstract

Romania was one of the first transition countries in Europe to introduce auctions for allocating

standing timber (stumpage) in public forests. In comparison with the former system in the

country--administrative allocation at set prices--timber auctions offer several potential

advantages: greater revenue generation for the government, a higher probability that tracts will

be allocated to the firms that value them most highly, and stronger incentives for technological

change within industry and efficiency gains in the public sector. Competition is the key to

realizing these advantages. Unfortunately, collusion among bidders often limits competition in

timber auctions, including in well-established market economies such as the United States. The

result is that tracts sell below their fair market value, which undermines the advantages of

auctions.

        This paper examines the Romanian auction system, with a focus on the use of

econometric methods to detect collusion. It begins by describing the historical development of

the system and the principal steps in the auction process. It then discusses the qualitative

impacts of various economic and institutional factors, including collusion, on winning bids in

different regions of the country. This discussion draws on information from a combination of

sources, including unstructured interviews conducted with government officials and company

representatives during 2003. Next, the paper summarizes key findings from the broader research

literature on auctions, with an emphasis on empirical studies that have developed econometric

methods for detecting collusion. It then presents an application of such methods to timber

auction data from two forest directorates in Romania, Neamt and Suceava. This application

confirms that data from Romanian timber auctions can be used to determine the likelihood of

collusion, and it suggests that collusion reduced winning bids in Suceava in 2002 and perhaps

also in Neamt. The paper concludes with a discussion of actions that the government can take to

reduce the incidence of collusion and minimize its impacts on auction outcomes.




                                                 2

                        DETECTING COLLUSION IN TIMBER AUCTIONS:
                                  AN APPLICATION TO ROMANIA



Introduction

Romania was one of the first transition countries in Europe to introduce auctions for allocating

harvesting rights for standing timber (stumpage) in public forests. The auctions involve small

tracts of forestland that must be harvested within a short period, typically less than a year from

the date of the sale. From this standpoint, the Romanian system is roughly similar to the system
used in U.S. national forests.1

         In comparison with the former system in the country--administrative allocation of timber

at set prices--timber auctions offer several important advantages in principle.                      From the

standpoint of the government, the most obvious one is that auctions are likely to generate more

revenues; in Romania; this was undoubtedly a primary impetus for introducing auctions. But

auction systems also have other potential advantages that may be more subtle but are arguably

more important. One is static efficiency in the forest sector: with auctions, tracts are more likely

to be allocated to the firms that value them the most. A second is dynamic efficiency: auctions

are better at creating incentives for technological change, which is the ultimate engine of cost

reductions in timber harvesting and wood processing; it will eventually allow firms to bid higher

amounts. A third is efficiency within the public sector. Auctions for timber in Romanian state

forests are administered by the National Forestry Administration (NFA). This is a financially

autonomous agency which, in addition to administering auctions, has responsibilities related to

forest management, regulation, and extension. It has 37 regional forest directorates (directia

silvica). Firms are allowed to participate in timber auctions in any forest directorate. This

choice creates an incentive for directorates to attract buyers by reducing bureaucratic hassles as

well as other pecuniary and nonpecuniary costs of doing business.

         As this last point implies, competition is the key to the realization of the advantages of


1It differs, however, from auction systems recently introduced in some other countries that are trying to create a

more market-based forest sector. An example is Cameroon, where auctions are being used to allocate long-term

(15-year, renewable once) timber concessions that cover large areas of forest.



                                                           3

auction systems. Unfortunately competition is often limited in timber auctions for a variety of

reasons. One that has attracted regular attention in the United States since the 1960s is collusion
(Mead 1967, Johnson 1979, Brannman 1996, Baldwin et al. 1997).2 Collusion is a coalition of

bidders who try to circumvent competition in order to obtain timber at a lower cost. Such

agreements are usually illegal, but detecting them can be difficult. If collusion is a problem with

timber auctions in the United States, an advanced market economy with well-developed legal

institutions, it is likely to also be a problem in transition economies like Romania. A previous

econometric study on auctions in the Suceava forest directorate found evidence that winning bids

were significantly lower for tracts with fewer bidders (HIID 1996). Although this is by no
means proof that bidders were colluding on those tracts,3 it does indicate the important role of

competition in determining the prices the NFA receives.

         This paper examines the Romanian auction system, and uses econometric methods to

detect collusion. The first and second sections describe the historical development of the system

and the main steps in the auction process. The third section describes the impact of economic

and institutional factors on winning bids in different regions of the country; it includes a brief

review of evidence on collusion.              Information in these first three sections comes from a

combination of sources, including unstructured interviews conducted with government officials,

company representatives, and NGOs during 2003. Individuals interviewed are listed in the

appendix. The fourth section summarizes some key findings from the research literature on

auctions, with an emphasis on empirical studies that developed econometric methods for

detecting collusion. The fifth section presents an application of such methods to recent timber

auction data from two Romanian forest directorates, Neamt and Suceava. Our results confirms

that data from Romanian timber auctions can be used to determine the likelihood of collusion,

and it suggests that collusion reduced winning bids in Suceava in 2002 and perhaps also in

Neamt, where less data were available. The sixth section concludes this paper.



2We review these studies in section 4.

3For example, tracts that sell for low prices might have few bidders not because of collusion but rather because they

are of low quality. To control for this endogeneity between the quality of a tract and the number of bidders, Vincent

(1996) used the predicted number of bidders rather than the actual number in a re-analysis of the data. He found less

evidence of an impact of bidder number on winning prices.



                                                           4

1.      Development of the timber auction system in Romania

During the Communist era, each county in Romania had a single timber seller, which was the

local NFA branch, and a single timber buyer, a state-owned forestry enterprise. The former sold

timber to the latter at prices set by the government. This situation remained relatively unchanged

for the first two years after Ceausescu was deposed. Then, in response to the emergence of

private enterprises in the forest sector and financial pressure, the NFA branch in Neamt decided

to experiment with auctions as a means of allocating standing timber to this new user group. The

NFA headquarters subsequently approved the use of auctions and encouraged other branches to

use them. Soon after, the government issued Ministerial Order 572 (1991), which mandated

auctions for selling timber to private enterprises. It did not, however, provide guidelines to

implement auctions. Each NFA branch introduced auctions at its own pace with its own rules.

        Government Decision 342 (1995) represented an initial attempt to create a consistent and

comprehensive legal basis for the auction system. Among other things, it extended the auction

requirement to include timber sales to state-owned enterprises. The Government had observed

that winning bids in auctions organized for private firms exceeded by a large margin the

administered prices paid by state-owned enterprises. Government Decision 695 (1998) brought

additional clarity and uniformity to the auction system. It mandated auctions in all public

forests, including the ones under the jurisdiction of local councils and not the NFA.                         It

authorized, but did not require, the NFA to administer auctions in non-NFA forests, including
private forests.4

        In addition to Government Decision 659, the current legal framework for timber auctions

includes Governmental Regulation 96 (1998) concerning the management and administration of

forests, together with the Forest Code. It also includes several detailed governmental and

ministerial regulations covering auction procedures, logging technologies and techniques,

monitoring and enforcement of logging operations (tree marking, legal documents for harvested



4In early 2004, Government Regulation 85 (2004) introduced new rules that require timber from forests controlled

by local councils to be included in the NFA auctions. It also included provisions to improve the advertising of

upcoming auctions and the accounting of logging companies' harvest quotas. See sections 2.5 and 2.6.




                                                        5

timber, control of timber flows during harvesting), and harvest contracts (payments for timber
harvested, payment and reimbursement of bid guarantees, resolution of disputes).5

         Relatively few modifications have been made to the framework laws and regulations in

the last five years, with the exception of regulations dealing with harvest contracts, which have

changed at least once a year. Ministerial Order 635 (2002) contains the main body of current

contract regulations. One important modification is NFA Order 6465 (October, 2001), which

mandated sealed bidding instead of oral auctions and prohibited sales of timber by negotiation.

The former mandate was quickly cancelled, but the latter prohibition remains in place. Another

important modification is Law No. 654 (2002), which authorizes long-term timber sales. We

discuss the motivation for this new law, and early experience with it, in Section 3.


2.       Steps in the timber auction process

In this section we describe the main steps in the timber auction process, in the approximate order

in which they occur.


    2.1.          Annual allowable cut

The process of selling timber starts with the determination of the annual allowable cut (AAC),

which is the maximum volume of timber that may be harvested in a given year. The building

blocks for the AAC are forest production units. These are relatively homogeneous areas of

forestland that are intended to produce a sustained and even flow of timber over time. Each

production unit is intended to eventually contain equal areas of each forest age class (a "normal

forest"). Every ten years, the National Institute for Forestry Research and Management Planning

(generally known by its Romanian acronym, ICAS), an autonomous unit within the NFA,
prepares management plans for the production units.6 These plans are reviewed by the Forest




5Governmental Regulation 96 (1998) was recently modified by Law No. 120 (2004), which mainly concerns private

forest administration and the enforcement of existing rules and regulations.

6 A few dozen private firms have begun preparing management plans in the past few years.    Most operate in

cooperation with ICAS instead of competing with it.



                                                            6

and Wildlife Territorial Inspectorates in the National Authority for Control.7 They contain

proposed annual harvest schedules for the production units over the ten-year period. These

harvest schedules are intended to ensure that harvests are in balance with timber growth. During

the last two decades, ICAS has proposed an aggregate annual harvest below the estimated

aggregate annual growth, to offset harvests that exceeded scheduled amounts in earlier periods.

         Each year, a top-down process is used to allocate the aggregate scheduled harvest across

regions of the country, categories of forest owners (the state, local councils, private parties, etc.),

and categories of forest users. This process establishes the actual AAC. There are three main

categories of users: "economic agents" (i.e., timber harvesting and wood processing firms), the

rural population, and the NFA itself. Two subcategories have been added to the economic agents

category since 2001: users who harvest for the needs of the National Construction Agency, and

users who have won long-term contracts. Before auctions, the NFA and holders of long-term

contracts select tracts to fill their user shares. The remaining tracts enter the auctions.

         The allocation of the AAC across regions and owner and user categories is significantly a
political act. It is proposed by the Ministry of Agriculture, Forests, Water, and Environment8

and must be approved by a parliamentary committee and signed by the Prime Minister.

Although different user groups understandably lobby to influence the allocation in their favor,

we found little evidence to suggest that corruption affects the allocation process.


    2.2.          Harvest scheduling and auction timing

At the level of forest production units, the AAC is implemented through the harvest schedule,

which indicates which tracts should be harvested in each year during the current decade. Tract

selection is driven largely by consideration of factors that affect long-run forest health, in

particular forest regeneration and forest protection. Natural catastrophes, such as windstorms,

forest diseases, and pest outbreaks, affect the actual implementation of the harvest schedule. The

removal of damaged timber is given higher priority than the harvesting of tracts on the original


7Before June 2003, the Inspectorates were in the Ministry of Agriculture, Forests, Waters, and Environment. They

also review forest management plans prepared by private firms.

8 Since March 2004, the ministry has been reconfigured as the Ministry of Agriculture, Forests and Rural

Development.



                                                          7

schedule, and when such salvage logging occurs, the harvest date for some of the tracts on the

original schedule is postponed to ensure that the AAC is not exceeded.

        Forest directorates organize two primary auctions each year. The first is in the spring,
traditionally between March 1st and April 15th, when 40 percent of the AAC for the current year
is sold. The second is in the autumn, traditionally between August 1st and September 15th, when

20 percent of the AAC for the current year and 40 percent of the AAC for the following year is

sold. Forest directorates can also organize secondary auctions for tracts that do not sell at the

primary auctions or for emergency cuttings. The managers of administrative subunits of the
forest directorates, which are termed forest divisions (ocolul silvic),9 have a lot of latitude in

allocating tracts between the spring and fall auctions.              NFA managers and staff at the forest

division level also decide the quantity of timber offered at secondary auctions as well as the

timing of those auctions.


    2.3.         Tree marking and timber inventory

The selection of trees to be harvested within a tract depends on the silvicultural treatment applied

(e.g., final harvest or thinning). Trees to be harvested are marked individually with a special

forest hammer. An exception is made for clear cuts: in that case, the tract boundaries are simply

marked in the field. Trees are inventoried one by one, with information collected on species,

trunk diameter 1.3 meters above the ground, and timber quality. Each tree is given a serial

number, which is painted on the trunk. Tree heights are measured for a sample of marked trees.

        Information from the inventory is recorded on the tract evaluation form (act de punere in

valoare, or APV). The NFA guarantees the timber volume reported on the APV to be accurate
within 5 percent.10 The APV constitutes the basis for timber transactions, because it identifies

the quantity and the quality of timber to be sold. The APV also plays an essential role in law

enforcement, because it provides proof that timber can be legally harvested. Each APV is given

a unique identifying number.




9The boundaries of forest divisions do not necessary coincide with the boundaries of forest production units.

10Winning bidders have a limited time in which to dispute the NFA's volume estimates and must incur what can be

considerable forest inventory costs in order to make their case.



                                                             8

    2.4.        Reserve prices

The government's reserve price, specified on the APV, is one of the few pieces of economic

information provided on the APV. It provides the starting point for bidding and is expressed in

lei per cubic meter, averaged across the total timber volume in a tract. When the government

issued the first comprehensive regulations for timber auctions in 1995, it declared that reserve

prices must be approved by the Ministry of Finance. In 1997 it removed reserve prices from the

list of state-approved prices and authorized the NFA to set reserve prices independently, guided

by two principles: covering the costs of forest administration, and setting a higher price for better

quality and more accessible timber. In April 2001, however, the government brought reserve

prices back under state supervision; currently, the NFA must submit proposed reserve prices to

the Competition Office for review. The reserve prices approved by the Competition Office must

be used in all auctions for public timber, though not in auctions for timber in private forests.

        Reserve prices are one of the most important tools that governments have to fight

collusion and ensure that the timber sold in auctions fetches a fair market price. Unfortunately,

the most thorough previous study on timber auctions in Romania, which analyzed 1992-95 data

from auctions in Suceava, concluded that "current starting [i.e., reserve] prices are far below

market values and inadequately reflect variations in market values, thus raising the risk that the

government is not receiving as much revenue as it could and reducing the ability of stumpage

prices to stimulate more efficient wood-based industries" (HIID 1996, p. 3). This is a result of

basing reserve prices on forest administration costs instead of on data related to the market value

of timber. We return to this point in the concluding section.


    2.5.        Marketing

Forest directorates are required to publish announcements of auctions at least 30 days before the

auction date in two newspapers, one of which must be local. Most directorates also publicize

auctions on the Internet. Several forest directorates--in particular, Arad, Ialomita, Mures, and

Timis--have aggressively implemented measures to attract more firms to participate in their

auctions. Examples include publicizing auctions a year in advance, promoting lesser-known

timber varieties, and meeting with representatives of harvesting firms to better understand their

needs. The Tomnatec forest division in Suceava, which until recently was managed by ICAS,




                                                   9

has been a pioneer in using direct marketing methods to publicize its auctions.         11  It faxes

invitations to firms and follows up with phone calls.

         Interested firms are allowed to review the APV for each tract before an auction. They are

also allowed to review a separate document that describes all the rules for harvesting the tract as

well as the deadline for signing the timber harvest contract by the winning bidder.


     2.6.         Qualification and registration of bidders and bid guarantee

Five days before an auction, the forest directorate decides which firms are qualified to

participate. There are four main qualification criteria: the firm must be legally constituted; it

must have timber harvesting as an economic activity; it must not have outstanding debts to the

forest directorate organizing the auction or to other forest directorates; and it must not have

exceeded its authorized timber harvest quota for the current year.

         The fourth criterion requires additional explanation. In 1995, as part of Government

Decision 695, the Government created a committee to determine the annual harvesting capacity

of individual companies in the forest sector. The committee has 9 members: 4 with expertise in

forest management, 4 with expertise in timber harvesting, and 1 with expertise in wood

processing. It reviews a company's technical capabilities, financial status, and human resources

and, on this basis, determines the total annual amount of timber that the company is authorized to

harvest. The NFA allows a company to participate in an auction only if the amount of timber the

company already has under contract is less than the company's annual quota. The NFA is

reportedly diligent in enforcing this rule.

         Firms that qualify to participate in an auction must register in advance for the individual

tracts they intend to bid on. For each such tract, they must pay the NFA a deposit, or bid

guarantee, which is equal to 5 or 10% of the estimated minimum value of a tract (= reserve price

times timber volume), depending on the size of the tract. The deposit is intended to discourage

frivolous bidding. The NFA returns the deposits to losing bidders within 3 days of the auction.




11It was transferred to NFA administration due to alleged mismanagement.



                                                          10

    2.7.           Bidding procedure

The NFA has experimented with both oral and sealed auctions. Until 2001, it used mainly oral

auctions. As noted earlier, in October of that year it prohibited oral auctions. Although this

prohibition was quickly cancelled, since January-February 2002 some forest directorates have
organized only sealed-bid auctions.12

        A sale is considered final only when there are at least two registered bidders and at least

one of them offers the reserve price. Tracts that fail to sell at the primary auction are then

proposed for sale at a secondary auction, which is typically scheduled within a few weeks of the

original auction. Until recently, the forest directorate was allowed to sell any tracts still unsold

after the second auction by negotiating with the last firms that expressed interest in them. A

verbal order of the NFA chief executive, issued in early 2003, prohibited such negotiated sales.


    2.8.           Timber harvest contract and harvest warranty

The high bidder must sign a timber harvest contract within 10 working days after the auction or

lose its bid guarantee. After signing, it must pay the NFA a deposit of 5 percent of the auction

value (= winning bid times timber volume) before it begins harvesting. This harvest warranty is

intended to discourage logging damage. The NFA returns the warranty at the end of the

harvesting operation if the remaining stand is not damaged.


    2.9.           Differences across forest directorates

Some of the procedures described above can differ across forest directorates. They include:

    · The rules for accepting the APV. In Neamt, firms cannot claim any adjustment for

        discrepancies between estimated and actual timber volumes once harvesting has begun.

        In Suceava, firms can request a re-evaluation of timber volumes (and qualities) up until

        the halfway point of the period scheduled for harvesting the tract.

    · The wording of harvest contracts, in particular rules related to breach of contract and

        litigation.

    · The possibility of negotiating deadlines related to the harvest contract.

    · The scheduling of payments.


12There is just a single round of bidding in the sealed-bid auctions.



                                                            11

3.      Effects of economic and institutional factors on bids in Romanian

        timber auctions

Bids in timber auctions are affected by both the physical characteristics of tracts, such as the

amounts and qualities of timber they contain or their terrain, broader economic and institutional

factors. The effects of physical characteristics are straightforward: for example, tracts with more

timber of higher quality tend to attract higher bids, while tracts with more difficult terrain tend to

attract lower bids. Differences in physical characteristics have been found to be important in

explaining bid variations across tracts in Romania (HIID 1996, Vincent 1996).

        In this section, we focus instead on the effects of various economic and institutional

factors, especially the ones that affect the degree of competition in timber auctions. These

factors can affect multiple auctions or multiple tracts within the same auction; they are more

"macro" than the "micro" characteristics of individual tracts. We begin with basic supply and

demand considerations and close with factors related to corruption and collusion. Our main

purpose is to understand the impact of these factors on the winning bids in timber auctions, in

order to provide a basis for formulating our econometric models and interpreting our results.


    3.1.       Regional timber markets within Romania

Geographical differences in forest composition, terrain, infrastructure, and other factors have led

to the development of several fairly distinct timber markets in Romania. These markets are by

no means autarkic--for example, officials in the Timis forest directorate tell of a wood

processing company from a town 400 km east that regularly purchases timber in the directorate's

auctions--but the differences are sufficiently great to produce persistent differences in the level

of winning bids. The very fact that winning bids vary across the regions suggests that timber

markets are functioning in the basic sense of reflecting differences in supply and demand forces.

        Table 1 shows data on scheduled sales of mature timber at auctions held during the first 7

months of 2003 in each forest directorate. Volumes are totals for the 7-month period, and

winning bids are weighted-average means (i.e., total sales values divided by total volumes sold).

"Mixed hardwoods" and "mixed softwoods" both refer to miscellaneous species in the broad-

leaved tree group. Mean winning bids that are below the national mean for a particular timber

group and directorate are shown in italics. The data in this table indicate that the directorates



                                                   12

vary greatly in terms of the amounts and types of timber sold by auction and the mean winning

bids for that timber. Suceava stands out as a directorate that sells a large volume of mature

timber, overwhelmingly coniferous, followed by Neamt. Bacau and Caras Severin are major

sellers of nonconiferous timber. Winning bids tend to be highest in directorates in the East,

Northwest/West, and South regions, and lowest in the Central, Danube, and Southwest regions.

         Based on this table and other sources of information, we can describe the regional timber

markets in the country as follows. The most timber-rich regions are the Northeast,

Northwest/West, East, and Southwest. The Northeast is the largest supply source for coniferous

species such as pine, fir, and spruce. It has a well-established industry with large capacity in

both timber harvesting and wood processing.

         Nonconiferous species are more important in the Northwest/West and East (oak and

cherry) and Southwest (beech). Among these three regions, demand tends to be greater in the

Northwest/West, due to high export demand from neighboring western countries.                               As a
consequence, winning bids have tended to be higher than in the East.13                        Demand in the

Southwest, which includes only the Caras Severin forest directorate, is very weak despite the

good quality of the region's beech stands. This region is lightly populated and has high wage

rates, due to a large mining industry and commercial trading opportunities with neighboring

Serbia. The tight labor market has discouraged the development of local timber harvesting and

wood processing industries.

         The Central region and the South have strong demand but poor supply. Factories in these

regions rely on timber harvested elsewhere. The Central region has the biggest investment in

wood processing in Romania, but it also has the smallest area of public forest in the country, and

forest ownership is divided among numerous small owners due to restitution. This depresses

demand for timber from local sources because mills prefer to deal with large, stable suppliers.

         The Danube region (counties of Bucuresti, Constanta, Dolj, Ialomita, Mehedinti, Olt,

Tulcea) is a special case. It has a substantial supply of poplar trees. There is some export

demand for poplar wood, but it is not consistent. There is also some domestic demand by the



13This was not the case during the first 7 months of 2003, however: winning bids were fairly consistently higher in

the East than in the Northwest/West.



                                                          13

pulp and paper industry. However, Law No. 570 (2003), adopted in late 2003, prohibits timber

harvests for 3 years in much of this region and parts of the South. The stated reason is to fight

desertification.


    3.2. Factors on the supply side

        3.2.1. Number of timber sellers

The NFA manages timber supply from state forests in Romania. These forests currently account

for about three-quarters of the country's forest area, but this proportion is declining due to

restitution. Nevertheless, the NFA remains the leader in the wood market in Romania. It sells

about four-fifths of the total volume of timber harvested in the country, and its supply decisions

have a strong influence on timber prices. Its ability to supply large timber volumes on a regular

basis makes it the preferred timber source for large firms.

        Local councils administer just above 10 percent of the total forest area, approximately

750,000 hectares. Of this amount, about 400,000 hectares are managed by the NFA on a contract

basis. Local councils sell standing timber by auction, using the same reserve prices as the NFA.

Timber from local council forests managed by the NFA is sold in the same auctions as timber

from the NFA's own forests (i.e., state forests). The amount of timber offered by local councils

is reportedly large enough in some locations to depress the winning bids in NFA auctions by a

significant amount. Local councils tend to put strong pressure on the Ministry to authorize a

larger AAC. They have chronically low budgets, and their primary forest management objective

is revenue generation.

        Individual private owners administer approximately another 10 percent of the total forest

area. They have differing management objectives. Most, like the local councils, have a strong

interest in short-term income. Others, however, would like to preserve the forest for a future

time when they expect to need wood for building houses, etc. Private owners are not required to

sell timber by auction, nor are they required to use the official reserve prices if they do (unless

the NFA administers the auctions on their behalf). At present, any influence they have on timber

markets tends to be very local and ephemeral.

        Owner associations administer forests in the eastern Carpathian Mountains.           These

associations trace back to common property forests that the Austrian Empire granted to local



                                                  14

communities in return for guarding its borders. Like private owners, the associations are not

required to sell timber by auction. They have a strong revenue motive and can consistently

influence prices in local markets.

        Churches (including monasteries) comprise the final category of timber sellers. Their

properties are individually small, 30 hectares each, and are spread all over the country. The total

area of their properties can be sizeable in counties with many churches, though still much less

than before the communist rule. For example, an estimated 15,000 hectares will be restored to

churches in Suceava (14,000 hectares have already been restored), compared to the nearly

200,000 hectares owned by them previously. The churches' main forest management objective

is to produce raw material for new buildings or restoration of existing ones. They tend to use

most of the timber they harvest. Their influence on the market is not significant in most forest

directorates except at a very local level.

        3.2.2. Annual allowable cut

The NFA and local councils are inelastic supply sources, owing to the determination of timber

harvests on the basis of AAC calculations instead of timber market conditions. The effect, well

known to forest economists, is to exacerbate the volatility of timber prices over the course of the

business cycle (Waggener 1969). When markets are strong, mills are starved for timber and bid
up prices to levels that can lead to widespread bankruptcies if markets cool off too rapidly.14

Conversely, when market conditions are weak, as they have been in Romania during much of the

post-communist period, the result is low bids: supply is too high relative to demand. Some tracts

might go unsold simply because they are put on the market at the wrong time, not because they

are inherently undesirable.

        To a certain extent, timber prices have been inadvertently supported by the policy of

setting scheduled harvests below growth and, since the early 2000s, by the failure of the Ministry
to set annual harvest quotas on time.15 Some auctions have been postponed or cancelled due to

the latter factor. Large salvage sales have partially offset this factor in some directorates (e.g.,

Neamt in 2002). Moreover, it is not clear that the policy to reduce scheduled harvests has been


14As happened in the U.S. in the early 1980s. See Mattey (1990).

15The Ministry did set the quota for 2005 on time.



                                                         15

followed in practice: the share of forestland in the 0-20 year age class, approximately 22 percent,

remains significantly larger than the share for any of the other five 20-year age classes (20-40,

40-60, 60-80, 80-100, 100+), even though communist rule ended 14 years ago.

        Planning by forest users--determining where the AAC is likely to bind and where it is

likely to make surplus timber available, and making wood-processing investments accordingly--

is difficult because ICAS's forest management plans are not publicly available. Potential bidders

do not learn about a tract's availability until an auction is advertised. This lack of information

can also be expected to contribute to increased price volatility, because milling capacity is less

likely to be in line with the supplies actually made available.

        3.2.3. Physical access to timber

A limited and poorly maintained road network increases the costs of forest access and thus

depresses bids in timber auctions. According to a report prepared for the World Bank by
Environmental Resources Management,16 Romania's road network includes 32,500 km of forest

roads, 7,600 km of public roads in or near forests that can be used for logging operations, and

1,450 km of roads that were constructed for other purposes, such as mining, but can be used for
logging.17 This network provides access to only 65 percent of Romania's production forests. Its

density, at 6.1 m/ha, is much less than in other European countries with mountainous forests.

For example, Austria, France, Germany, and Switzerland all have 25-45 m/ha of roads in regions

with timber producing forests.

        Most of the forest roads in Romania were constructed before 1985. They became the

responsibility of the NFA in 1994. Although the NFA has constructed more than 300 km of

forest roads and rehabilitated nearly 300 km, some 7,000 km have become unusable due to

deterioration and inadequate maintenance. Effective access is thus even less than the figures in

the previous paragraph would indicate.




16Environmental Assessment Update, Volume I: Summary and Main Report (September 2002), prepared for the

Romania Forest Development Project.

17There are also 65 km of forest railroads.



                                                   16

    3.3. Factors on the demand side

        3.3.1. Number of timber buyers

Only domestic firms are permitted to participate in the NFA's auctions for standing timber.

Winning firms are not allowed to resell tracts to other firms, domestic or foreign. The NFA

formerly organized a separate set of auctions for harvested logs--not standing timber--that were

open to foreign firms, but it suspended them following strong lobbying by domestic wood

processors. Foreign demand thus influences Romanian stumpage prices only indirectly, through

derived demand resulting from exports of sawn wood and other processed products. Though

indirect, this influence is significant. Economic conditions in the Middle East, which is the

traditional import leader for Romanian sawn wood, are particularly influential. The stumpage

price for beech fell after Middle Eastern demand for steamed beech sawn wood fell. Economic

conditions in Western Europe are becoming more influential. Exports to the west have increased

recently due to Western European investments in the wood processing industry.

        Domestic timber buyers fall into two categories. The first includes free-standing logging

companies that purchase standing timber and sell the harvested logs to wood processing
companies. According to a report by WWF,18 the original single, state-owned logging company

split into three regional companies in 1993. By 1994, these had split into 17 companies with

majority state ownership. By 1996, they had split again into 32 companies. The state remained

the major shareholder, but the proportion of private shareholders increased. This trend of rising

private ownership has continued. In addition to the splitting and privatization of former state-

owned companies, around 1,000 small private logging companies emerged during the 1990s.

Many have since gone bankrupt and ceased operating. Around 600-700 remained as of 1999.

        The second category includes companies that process wood. According to the National

Institute of Statistics, in 2002 there were 8,634 commercial enterprises in the wood-processing

sector (including furniture and pulp & paper), of which 271 were classified as large (more than

250 employees). Most of the large firms have mixed state/private ownership. Of the remaining

enterprises, 1,589 were classified as medium (100-250 employees) and 6,774 as small (less than


18Technical report on the implementation of the "Certified Forest Products Trade Analysis and Facilitation" Project

in the countries of Slovakia and Romania, prepared by the WWF-Danube Carpathian Programme for the Global

Forest and Trade Initiative (undated, but evidently mid-2000).



                                                          17

100 employees).19 Most of the small and medium firms are privately owned.20 The small

enterprises are mainly primary processors--sawmills, veneer mills, plywood mills--while most

of the medium and large ones go further downstream and make furniture, parquet, etc.

        Some of the wood-processing firms are vertically integrated upstream and have their own
logging crews. According to the same source,21 in 2002 there were 2,246 companies engaged in

timber harvesting in the country, including 43 territorial units of the NFA. Of these, 1,939 were

limited liability companies, which correspond to small and medium firms, while 81 were joint

stock companies, which correspond to large firms. Sixty-one percent of the 1,692 companies

that completed the government's annual questionnaire stated that they processed wood in

addition to harvesting it. Sawmills are especially likely to have their own logging crews, and

large sawmills are especially aggressive in seeking cheap raw material. They use their ability to

buy large amounts on a regular basis to convince individual private owners and forest

associations to sell standing timber to them outside the auction system and at attractive prices.

        There is thus a large pool of timber buyers in the country, albeit one that is relatively

shallow in terms of larger companies. In the absence of local monopsonies or collusion (see

section 3.4.2 below), this large pool should lead to highly competitive auctions. One potential

entry barrier is the central commission that authorizes firms' harvesting capacities, but this

barrier is porous. The commission's evaluation is based on documents provided by firms. The

commission does not check the actual equipment and other resources that firms claim to own or

have access to for harvesting. This creates an incentive for firms to exaggerate their harvesting

capacity in order to receive a larger authorization and thus qualify to buy more timber.




19There were no small- or medium-scale wood-processing companies (including pulp and paper and furniture)

before 1990, when the existing 244 large companies in the country were split to improve their management and to

make them more attractive for privatization.

20Some churches also operate mills as a money-making venture (not just to produce lumber for their own use).

21"Ancheta statistica privind volumul exploatat de catre agentii economici atestati si suprafata parcursa cu taieri in

anul 2002." Bucuresti, 22 p.




                                                          18

          3.3.2. Technology

Technologies continue to be outdated and production processes inefficient for many firms in the

Romanian forest sector. This is especially true for timber harvesting companies. Inefficient

technologies reduce the amount firms can pay for a cubic meter of standing timber.

          The small size of the tracts auctioned by the NFA has been identified as one obstacle to

more rapid technological change in the sector. A single tract is sufficient to supply a company

for only a short period. Even if a company succeeds in purchasing many tracts at one auction, it

must harvest them all within a half-year or so. The resulting lack of resource security reportedly

diminishes the incentive to invest in improved harvesting and processing equipment.

          In response to this situation, in 2002 the Government issued Law No. 71 (operational in

2003): it authorizes auctions for 10-year timber supplies, but only firms with wood-processing

facilities are allowed to participate. Initial experience has been disappointing as none of the

large foreign investors in the Romanian wood-processing sector (e.g., Ambro, Fratti or
Schweighoffer) has participated so far. 22 They had evidently no interested in forming long-term

partnerships with Romanian firms in order to qualify to bid on the sales. This is discouraging

because foreign investors have been primarily responsible for the limited technological change

that occurred in the sector, such as the capability to process small diameter logs.


     3.4. Corruption and collusion

          3.4.1. Corruption

Quantitative measures of the amount of corruption in Romanian timber markets do not exist.

Allegations of corruption come mainly from logging companies. They fall into two categories,

bribery and fraud. Companies claim that NFA officers have sometimes demanded bribes in

return for approving one or more of the numerous requirements that firms must satisfy to


22One million cubic meters were offered for sale in 2002. Although 60 companies registered, they bought only

200,000 cubic meters. A smaller amount of timber, 800,000 cubic meters, was offered in 2003. Although nearly all

of it sold, there were fewer bidders than in 2002, and they did not bid vigorously. Most of the contracts sold for just

two bidding steps above the reserve price. Demand remained weak in 2004, when 400,000 cubic meters was

offered. This Information is drawn from editorials posted on the website, forestry.ro.




                                                            19

participate in auctions or to harvest the timber they have purchased.23 These claims appear to be

more common in regions where firms are small and have few alternative sources of timber.

When we asked a representative of one firm what would happen if he refused to pay bribes, he

answered that he would never refuse, because forest officers wield too much discretionary

power: they determine which roads to repair, thus affecting the firm's access its tracts; which

tracts to guard, thus protecting the firm's timber and logging equipment from theft; and what

penalties to assess in case of timber harvesting delays or logging damage, thus determining the

amount of the harvest warranty returned to the firm.

        The second alleged form of corruption involves NFA officers selling timber outside the

auction system. One story we were told is that an NFA forester might mark some trees only at

the base of the trunk and then cover the marks with leaves and branches, thus causing the trees

not to be recorded on the APV as part of the sale. The forester can then sell this unrecorded

timber to the firm that wins the tract and pocket the payment himself.

        To the extent they exist, these two forms of corruption can be expected to have opposite

impacts on firms' bidding behavior. If firms expect that they will need to pay bribes in order to

obtain necessary approvals, then they will place lower bids. On the other hand, if firms expect

that they will be able to harvest more timber than the APV indicates is present in a tract, then

they may place bids that appear unusually high. Information on corruption in the sector is too

anecdotal to speculate on which impact is likely to dominate.

        3.4.2. Collusion

        Romania's history of having one large state-owned harvesting enterprise in each forest

directorate, combined with the slow emergence of large privately-owned new enterprises, has

resulted in the domination of timber markets by one or a few large firms in most forest

directorates. A substantial portion of the national harvest is made by just three companies,

Forestar, Euroforest, and Romanel.              In some directorates, dominant firms reportedly avoid

competing with each other by dividing the directorate into regions of influence where they act as

local monopsonies. This can be expected to result in excessively low winning bids.




23Bribes are reportedly often paid in kind (a truckload of logs) instead of in cash.



                                                            20

         The opposite, excessively high bids, can occur if dominant firms engage in preclusive

bidding to defend their turf. This reportedly has also occurred. Because large companies buy

many tracts, their average purchase price is not significantly affected if they overbid on a few
tracts in order to prevent other companies from winning.24 An extreme example occurred in

1999, when one large company essentially prevented an auction from occurring. Few other firms

even dared to enter the auction hall. The company intimidated those who did by placing opening

bids two to three times greater than the reserve price.

         If collusion is occurring in Romanian timber auctions, then the former effect--

excessively low winning bids--should be more common than the latter. Dominant firms can be

expected to use preclusive bidding sparingly, to avoid dissipating the benefits of collusion. If the

costs of excluding competitors or enforcing a cartel are too high, then it does not pay to collude.

For these reasons, our econometric analysis of collusion in Romanian timber auctions focuses on

detecting excessively low winning bids. In view of the points discussed in sections 3.1-3.3, this

analysis obviously must control for other factors that influence the level of winning bids. Before

providing more detail on our analysis, we consider methods that other researchers have applied

to detect collusion. Our analysis uses similar methods.


4.       Previous research on collusion

Interest in methods for detecting collusion has increased during the past twenty years, in tandem

with a rising interest in using auctions to allocate a broad range of resources from the public to

the private sector. Several recent papers in the research literature have employed econometric

approaches to detect collusion. In this section, we summarize some of the main findings from

this literature.

         At the outset, it is important to emphasize that there is no universal strategy for detecting

collusion using auction data (Hendricks and Porter 1989).                     Instead, institutions, market

characteristics, and bidder heterogeneity need to be taken into account in tailored ways in

empirical studies. Most of the empirical studies have relied only loosely on auction theory in


24Some small companies say they are "slaves" of larger companies, unable to compete within them in the auctions

and dependent on their trading chains. On the other hand, some large companies complain about unfair competition

from small companies, which they claim evade taxes and thus can afford to bid more for timber.



                                                          21

determining how to deal with these complexities. In part this is because the theory of collusion
in auction markets is surprisingly recent.25 Even today, most of the theory pertains to simple,

one-time auctions in which just a single item is offered for sale. A well-developed theory of

collusion in repeated auctions in which multiple items are offered for sale--that is, a theory that

matches two of the basic characteristics of timber auctions--does not yet exist.


    4.1.          Early studies on collusion in timber auctions

Timber auctions have been a recurrent focus of empirical auction research. Anecdotal evidence

of collusion in auctions for timber in U.S. national forests has repeatedly been documented (e.g.,

see U.S. House of Representatives 1977), although few cases have been successfully prosecuted.

Early studies focused on differences between bids in oral and sealed-bid auctions as a means of

inferring the presence of collusion.

         The first study was by Mead (1967). He began by considering how basic industry

characteristics affect the merits of oral versus sealed-bid auctions, leaving aside the risk of

collusion. He argued that oral bidding is more appropriate when several of the following

conditions hold: fixed investment is high, transportation costs are important, sources of raw

materials are few, specific qualities are needed, sales are frequent, and financing is limited. He

claimed that these considerations explain why oral auctions are used more frequently for timber,

whereas sealed-bid auctions are typically preferred for oil and gas. Mead then went on to argue

that oral auctions are more susceptible to collusion because they facilitate preclusive bidding.

Many years later, Robinson (1985) provided theoretical support for this claim in the context of a

single-object auction: assuming that cartel members share the same information, he proved that

cartels are stable for oral ascending-bid auctions but not for sealed-bid auctions, in both common
value and independent private value auction models.26 In the empirical portion of his paper,

Mead presented evidence indicating that oral bids had indeed generated lower prices in Forest


25For example, Milgrom and Weber (1982, p. 1090) stated that "Situations in which bidders collude have received

no attention in theoretical [auction] studies."

26In a common value model, the object for sale is assumed to have the same value for all bidders, but bidders do not

know this value in advance. They differ in terms of their estimates of the object's value. In an independent private

values model, the value of the object differs across bidders, and bidders know their own (but not competitors')

values.



                                                          22

Service timber sales than sealed bids (the Forest Service uses both bidding procedures), and he

concluded that collusion was the reason.

        Johnson (1979) contested this conclusion, offering an alternative explanation related to

differences between firms. He noted that theoretical results in Vickrey (1961) indicated that oral

and sealed bidding should yield similar prices on average when firms are homogeneous and

bidding is competitive (no collusion). When firms are heterogeneous, however--for example,

when some bidders have a known comparative cost advantage over others--Johnson found that

oral bidding can yield lower prices even when bidding is competitive. Further muddying the

water, a subsequent theoretical study by Milgrom and Weber (1982) demonstrated that the

opposite result--winning prices in oral auctions exceeding those in sealed-bid auctions--can

occur in competitive auctions when bidders are symmetric and risk neutral and have "affiliated"
values for the object being auctioned.27 The upshot is that a simple comparison of winning

prices in oral and sealed-bid auctions does not provide a reliable means of detecting collusion.


    4.2.         Testing for collusion using econometric methods

More recent empirical studies on collusion have used increasingly refined statistical techniques.

They have examined auctions for not only timber (e.g., Brannman 1996 and Baldwin et al. 1997)

but also various other goods and services, including offshore oil leases (Hendricks and Porter

1988), construction contracts (Porter and Zona 1993, Bajari and Ye 2001), and school milk

contracts (Porter and Zona 1999, Pesendorfer 2000). Porter and Zona (1993) emphasized that a

simple and universal procedure for detecting collusion does not exist and is unlikely to be

developed. In large part this is because collusion can take many different forms, including

kickbacks (a cartel member pays others to win a specific auction), bid rotation schemes (cartel

members alternate who wins), and phantom bidding (cartel members submit fake, losing-bids to

create the appearance of competition). In fact, if a universal test for detecting collusion were

available, a knowledgeable cartel would have an incentive to devise a bidding strategy, based for

example on phantom bids, such that the auction data would appear to have been generated by a

competitive process.



27"Roughly, [affiliation] means that a high value of one bidder's estimate makes high values of the others' estimates

more likely" (Milgrom and Weber 1982, p. 1096).



                                                           23

        The inherent unobservability of key variables is also a problem. The ultimate goal of

collusion is to secure high profits, but data on bidders' profits are typically difficult to obtain.

State-of-the-art econometric procedures for detecting collusion therefore are linked instead to

observable aspects of bidder behavior that are expected to differ in competitive and collusive

contexts. These procedures attempt to go beyond the simple detection of patterns that are

consistent with collusion, since such patterns could instead merely reflect unobservable

characteristics of the bidders or the items being auctioned. In most cases, the procedures involve

estimating at least two models, with the structure of one being consistent with bidding being

competitive while the structure of the other is consistent with bidding being collusive. The two

models are then compared in terms of how well they fit the data.

        We review here the procedures employed in three empirical studies: Porter and Zona

(1993), which examined auctions for state highway construction contracts on Long Island, and

Brannman (1996) and Baldwin et al. (1997), which like Mead (1967) and Johnson (1979)

examined U.S. Forest Service timber sales. All three concluded that collusion was occurring.

Porter and Zona tested for collusion by analyzing differences in the ranking of competitive and

cartel bids using an ordered logit model. With one big exception, they had relatively little

information about the firms. The exception was that they knew the composition of the cartel at

the start of the sample period. They knew little about the jobs put out for bid aside from the

winning bids. Their explanatory variables included a dummy variable for whether a firm is

headquartered on Long Island and a "Backlog" variable that gives the dollar value of contracts

won and not yet completed by a firm. Other variables are built from "Backlog," such as "Cap,"

defined as the maximum backlog over the periods considered, or "Util" (= "Backlog"/"Cap"),

which accounts for how busy a firm is. Porter and Zona found that the ranking of cartel bids did

not coincide with the ranking of costs and thus rejected the null hypothesis of no phantom bids,

an outcome which suggests the presence of collusion.

        Brannman (1996) focused on the effect of potential competition on winning bids.

According to economic theory, potential entrants to an imperfectly competitive market should

limit the ability of participating firms to collude. Brannman proposed two measures of potential

competition. The first one, called simply potential competition, is a count of the number of firms

whose closest mill lies within the boundaries of the market for Forest Service sales. The second

one, called expected competition, is the sum of the estimated participation probabilities of firms


                                                    24

capable of taking part in an auction. He estimated these probabilities using a probit analysis of

the link between auction participation and hauling distance.

        Brannman included five categories of explanatory variables in his regression models: in

addition to the competition variables, Forest Service appraisal components, sale characteristics,

bidder characteristics, and product market information. Appraisal components included total

appraisal value, selling value, harvesting costs, and manufacturing costs, all per thousand board

feet, in addition to the average hauling distance to rival mills and its standard deviation (both

were excluded if the expected number of bidders was included). Observed sale characteristics

included tract acreage, percentage per-acre material, a dummy for salvage value, road

construction costs, a county dummy, and whether the sale was a Small Business Administration

set-aside sale. Bidder characteristics had two components: hauling distance to the winner's

closest mill, and whether or not a firm is large. Finally, product market information was

summarized by average monthly lumber selling prices and contract duration. One difference

compared to Porter and Zona is that Brannman did not account for the contracts the firms already

held.

        Brannman analyzed data from 1977 for both oral and sealed-bid Forest Service auctions.

He found that both measures of potential competition were important in explaining winning bids,

but not as important as actual competition (ie., the actual number of bidders). He argued that this

suggests the presence of collusion. In addition, he found that longer hauling distances were

associated with higher winning bids in the sealed-bid auctions. Echoing Mead (1967), he offered

preclusive bidding as a likely explanation.

        Unlike Brannman, Baldwin et al. (1997) focused exclusively on winning bids in oral

auctions. They also based their analysis more directly on the theory of bidder behavior, in an
independent private values auction framework.28 Brannman had argued that timber auctions fit

neither the common value model nor the independent private values model. The key feature of

their study was that it used a model presented in Graham and Marshall (1987) to endogenize the

effective size of a bidder cartel. The authors had no direct information on whether a cartel

existed among Forest Service timber buyers or, if it did, how many firms it contained. They only



28Vincent (1996) also used an independent private values framework in his analysis of bidding behavior in Suceava.



                                                          25

knew the number of bidders who registered to bid, the highest and second-highest bids, and the

identities of the firms who placed those bids. Using assumptions about the distribution of the

unobserved private values of the bidders, they developed a maximum likelihood procedure that

predicted the size of the cartel that best explained the observed winning bids, controlling for such

factors as the volume of timber in a tract, timber quality, logging costs, the terms of the harvest

contract, local inventories of timber, and special events that took place during the study period

(e.g., natural disasters). They then compared the likelihood statistic from this collusive model to

the likelihood statistic from a noncooperative model that included the same variables but no

allowance for endogenous cartel formation.

        Baldwin et al. analyzed a relatively small sample of 114 Forest Service sales that took

place between 1975 and 1981, out of a total of 3,355 sales. They limited the sample so severely

in order to ensure that the data were as consistent as possible with the assumptions of the

independent private values framework. They estimated five models: a non-cooperative (i.e.,

competitive) model that treated the sale of each timber tract as a single-unit auction event; a

single-unit collusive model; a non-cooperative, multi-unit model that included supply effects

resulting from multiple tracts being offered at the same auction (such supply effects could

confound the detection of collusion); and two nesting models that combined collusion with the

multi-unit model. Their multi-unit model was based on Weber (1983) and Milgrom and Weber

(1982), and it was an innovation in the empirical auction literature. They concluded from their

econometric results that winning bids were best explained by the models that allowed for

collusive behavior.


5.      Econometric analysis of collusion in Romanian timber auctions

This section presents our econometric analysis of collusion in Romanian timber auctions. Like

the mid-1990s study on Romanian timber auctions described in HIID (1996) and Vincent (1996),

we analyze data from Suceava, but we also analyzes data from Neamt. The auctions in both

cases are oral (ascending-bid) auctions. Neamt was the first forest directorate in Romania to

introduce timber auctions as a regular means of selling standing timber to both private and public

enterprises, having done so shortly before Suceava. Romanian forestry experts interviewed

during June-July 2003 claimed that large enterprises commonly exert market power in Neamt.




                                                   26

Such concerns were less strongly expressed about auctions in Suceava. This difference makes

for a potentially interesting comparison of the two regions.

        The biggest difference compared to the mid-1990s study is that we examine the issue of

collusion directly. The earlier study focused on determining whether reserve prices reflected

differences in the value of tracts put up for bid. In the case of Suceava, we apply the competitive

and collusive single-unit models developed by Baldwin et al. (1997). In the case of Neamt, we

apply a cruder approach, due to a lack of data on the number of registered bidders for each tract.

        The need to apply a cruder approach in Neamt points to another difference compared to

the earlier study: available data on tract characteristics were less detailed, even for Suceava. Few

data related to the auctions are computerized in either forest directorate. The field team for the

earlier study had more time and resources to extract relevant data from written records and enter

them into electronic databases. We were able to compile data on the volume of timber in a tract

and the type of timber sale (mature, thinning, salvage) for both forest directorates, and several

additional tract characteristics for Suceava. This amount of data on tract characteristics is not

unsually small compared to the studies on U.S. timber auctions reviewed in section 4, so at least

we can say that we controlled for the effects of tract characteristics on winning bids as well as

those studies.

        We saw in section 3 that many factors other than collusion and tract characteristics can

affect winning bids. We must control for those factors in order to isolate the effects of collusion.

The fact that regional timber markets are fairly distinct in Romania is one reason, in addition to

differences in data availability, that we analyze Neamt and Suceava separately. Due to the lack

of data, for the most part we were unable to develop variables directly related to the other

economic and institutional factors. Instead, we used dummy (0-1) variables to control for them.

Several of the factors discussed in section 3 can be expected to vary substantially within a forest

directorate: number of timber sellers other than the NFA, number of timber buyers, and forest

access. To capture the effects of these factors, we included dummy variables for the forest

divisions within the directorates. To the extent that corruption is present at different levels in

different forest divisions (perhaps absent altogether in some or many), the same dummies will

capture its effects. Because many of these factors have changed over time, we also included

dummy variables for the dates when the auctions occurred. The time dimension is especially



                                                    27

important for other factors, such as the AAC and the level of technology in the industry. While

the use of dummy variables is fine for the purpose of controlling for potentially confounding

factors in order to isolate the effects of collusion, it does not allow us to estimate the impact of

those other factors on winning bids. The dummy variables are standing in for several "real"

variables simultaneously.


    5.1.        Econometric analysis of data from Neamt

        5.1.1. Background

Neamt is a forest directorate in northeastern Romania just south of Suceava. The IRIS field team

obtained data on all timber tracts sold in the directorate between May 1998 and June 2003. The

data set included tracts that sold at negotiated prices in addition to tracts that sold by auction.

The former category evidently includes mainly tracts that were originally offered for sale at

primary or secondary auctions but failed to sell. The total of number of tracts in the data set was

4,513. Of these, nearly three quarters--3,355 tracts--sold by auction. Auctions have thus been

the main mechanism for selling timber tracts in Neamt.

        Table 2 presents basic characteristics of the auction sales during 1999-2003. Note that

the data for 2003 cover only the period January-June. Scheduled sales of mature timber and

thinnings accounted for most of the sales in terms of both volume and value in all years except

2002, when windstorms resulted in unusually large salvage sales. Scheduled sales were reduced

in that year to maintain harvests within the AAC. Mature timber accounted for most of the

scheduled sales in every year, especially in value terms.

        The bottom panel of the table shows mean winning bids per cubic meter, weighted by

timber volumes. Like the values for total revenue, these values are expressed in nominal terms.

Inflation thus accounts at least partially for their tendency to increase over time for most sales

categories. Within a given year, the pattern of values is generally what one would expect if bids

reflect relative timber values: for example, mean winning bids are higher for scheduled sales of

mature timber than for scheduled sales of thinnings. Auctions in Neamt at least satisfy this basic

criterion of allocative efficiency.

        Table 3 presents the same information for timber that sold at negotiated prices. Data

were available only through the end of 2002. To begin, note that the volume and especially the



                                                   28

value of timber sold by negotiation were much less than the corresponding amounts sold by

auction. This was true even before October 2001, when NFA Executive Order 6465 banned

negotiated sales. Scheduled sales accounted for a smaller proportion of negotiated sales than

auction sales in all years except the windstorm year of 2002. Mean negotiated prices were

typically much less than mean winning bids from auctions. All of this is consistent with the

tracts sold by negotiation being the less desirable ones.

        Like most previous studies of timber auctions, we focused our econometric analysis on
scheduled sales of mature timber.29 Such tracts accounted for only 869 of the tracts in the Neamt

data set, but as the amounts in Tables 2 and 3 indicate they accounted for most of the combined

value of timber sold by auction and negotiations during 1999-2002. The volume of timber in

these tracts ranged from 5 to 8,357 cubic meters, with a mean of 1,765 cubic meters. Winning

bids ranged from 55,000 to 1,577,000 lei per cubic meter, with a mean of 478,237 lei per cubic

meter. The ratio of winning bid to reserve price had a mean value of 1.4. Some tracts sold for as

much as 4.6 times the reserve price, which suggests that competition among bidders was at times

intense.

        The amount of data available for explaining the variation in winning bids was limited.

This is because very little auction data is computerized in Neamt compared to Suceava. The

IRIS field team was able to obtain only the following basic data on each tract:

    · Name of forest division (ocolul silvic) where the tract was located

    · Volume of timber in the tract

    · Type of timber (mature, thinning, salvage)

    · Auction date

    · Reserve price

    · Winning bid

    · Name of winning bidder




29Estimating the amount of harvestable timber and the cost of harvesting are more difficult for salvage sales and

thinnings than for sales of mature timber. Consequently, forestry departments and logging companies have a harder

time formulating reserve prices and bids, and econometric models of bidding behavior perform more poorly.



                                                          29

        The team obtained a small amount of additional information on winning bidders (e.g.,

addresses), but formatting inconsistencies precluded its use. It did not obtain information on the

number of companies that registered to bid on a tract, the number that actually bid, or the bids

they placed, other than the winning bid.

        Due to these data limitations, our test for collusion is much weaker for Neamt than for

Suceava. We explain this test in the next section. We close this section by noting two reasons to

believe collusion could be common in Neamt. First, individuals interviewed by the IRIS field

team in June and July 2003 reported that it is. They pointed to the fact that just three firms--

Petroforest, Romanel Forest Star Tarcau, and New Star--harvest more than half of the total
amount of timber sold in the directorate.30 These firms are somewhat less dominant in the case

of scheduled sales of mature timber, however, accounting for just over a third of the total amount

sold by auction.

        Second, and perhaps more convincing, winning bids were barely above the reserve price

for many tracts. A visual examination of the cumulative distribution of the ratio of winning bid

to reserve price revealed a break point at 1.07. Almost exactly half (436 out of 869) of the

mature timber tracts sold by auction in Neamt during 1998-2003 had ratios below this value.

The mean winning bid for these tracts, 349,780 lei per cubic meter, was significantly different
from the mean winning bid for the tracts with a higher ratio, 607,585 lei per cubic meter.31

Hence, lower ratios coincided with lower winning bids. They were not a consequence of, say,

reserve prices coincidentally being closer to winning bids for more valuable tracts.

        5.1.2. Theoretical model and econometric approach

Following Baldwin et al. (1997), we define:

                  Bi      Winning bid for tract i (lei)

                  ri      Reserve price for tract i (lei)

                  vi      Volume of timber in tract i (cubic meters).




30These firms also buy timber in neighoring forest directorates.

31The means were statistically different regardless of whether we assumed the variances for the two groups were the

same or different.



                                                           30

We distinguish the observed winning bid per cubic meter, Bi/vi, from the winning bidder's
valuation of the tract: the bidder's maximum willingness to pay.32 We assume that the valuation

has a lognormal distribution with mean zi and variance 2, where

                 zi       Vector of covariates that affect valuation of tract i.

We included as covariates the volume of timber, vi, and the volume squared; the natural

logarithm of reserve price per cubic meter, ln(ri/vi); and dummy variables for forest division and

auction date. We included the reserve price in view of the limited data on tract characteristics.

As noted in section 2.4, reserve prices are supposed to reflect differences in timber quality and

accessibility. The forest division dummies might capture these differences, too. The dummies

for auction date control for inflation, market conditions, and differences in auction procedures.

         Our testing procedure amounted to evaluating the relative performance of a reference

model in which winning bids are assumed to equal bidders' valuations for all tracts to one in

which winning bids are assumed to be below valuations for some tracts as a result of collusion.

More formally, the null hypothesis is that all winning bids, including those close to reserve

prices, equal actual valuations:

                                             ln(Bi/vi) = zi + i  i.                                             (1)


The alternative hypothesis is that only winning bids that are not close to reserve prices equal

actual valuations. Winning bids that are close to reserve prices instead equal lower bounds on

actual valuations:

                                       ln(Bi/vi) = zi + i if Bi/ri  1+                                          (2)

                                      ln(Bi/vi)  zi + i if Bi/ri < 1+,

where  = 0.07. We test the null by comparing the log likelihood statistic from model (1), which

is a simple linear regression model, to the log likelihood statistic from model (2), which is a

censored regression model. If the absolute value of the log likelihood statistic from model (2) is

smaller, then we reject the null and conclude that collusion could be present. As described in



32Strictly speaking, the winning bid equals the sum of the actual valuation by the second highest bidder plus the

additional increment offered by the highest bidder.



                                                         31

more detail in a moment, our analysis does not rule out explanations other than collusion. If

instead we fail to reject the null, then we fail to detect evidence of collusion.

        As emphasized, rejection of the null does not necessarily imply that collusion is present.

Collusion is a type of unobserved heterogeneity: collusion affects some tracts but not all, and we

cannot directly observe which ones it affects. Other tract characteristics have this feature, and

they offer competing explanations why model (2) might outperform model (1). One such

characteristic is the order in which tracts are auctioned. This is a characteristic that in principle

is easily observable, but we have no data on it for the auctions in Neamt. So, from our

standpoint as econometricians, it is unobserved. To see how it might generate a set of winning

bids that look the set that results from collusion, consider the following simple example.

Suppose n bidders attend an auction, and m identical tracts are offered for sale. The number of

tracts equals the number of bidders: n = m. Suppose that, unknown to other bidders, each bidder

demands only a single tract and is willing to pay the same maximum amount for it. The high

bids for the first m-1 tracts in this auction will equal the maximum willingness to pay (the high
bidders' actual valuations), but the high bid for the mth tract will equal the reserve price, because

the sole remaining bidder has no competitors. Model (2) would fit the data from this auction

better than model (1), but the reason would be diminished competition for the final tract, not

collusion.

        5.1.3. Results

Results are shown for Model (1) in Table 4 and Model (2) in Table 5. The coefficients on the

timber volume variables and reserve price are highly significant and similar in magnitude in both

tables. The positive sign on timber volume indicates increasing returns to scale in logging, while

the negative sign on the squared value of this variable indicates that the scale economies

diminish with the volume harvested.         The coefficient on reserve price is not significantly

different from 1. Given that both this variable and the dependent variable are expressed in

logarithmic form, the implication is that winning bid is proportional to reserve price, after

controlling for timber volume, forest division, and auction date. The coefficients on many of the

dummy variables for forest division and auction date, which are not shown, were also

significantly different from zero.




                                                     32

        The absolute value of the log likelihood statistic is substantially smaller in Table 5 than in

Table 4. The censored regression model, in which we assume that actual valuations exceed

observed winning bids for tracts that sell close to the reserve price, thus fits the data better than

the standard linear regression model. We therefore cannot rule out the presence of collusion. We

investigated the robustness of this result by re-estimating Model (2) with  set equal to values set

slightly above and below 0.07: respectively, 0.10, which increased the number of censored

observations to 449; and 0.06, which decreased the number to 435. The absolute values of the

log likelihood statistics were both larger than the value in Table 5.33 The cutoff value of  that

demarcates tracts that have abnormally low winning bids in view of their timber volumes,

reserve prices, locations, and auction dates does indeed appear to be 0.07.

        We estimated just how low the winning bids for these tracts might be by using the model

in Table 5 to predict the uncensored values for the tracts. The mean value across the 436 tracts

was 669,302 lei per cubic meter, which is much larger than the mean of the actual winning bids

for the tracts, 349,780 lei per cubic meter, but not significantly different from the mean winning

bid for the 433 uncensored tracts, 607,585 lei per cubic meter.

        5.1.4. Conclusions: Neamt

We conclude that collusion could be affecting the winning bids in many of the tracts sold in

Neamt timber auctions by a large amount. Although our results are strongly suggestive, we

cannot rule out other factors as explanations for the unusually low bids for these tracts.

Developing a more definitive test of collusion requires a richer data set, such as the one from

Suceava. We turn to that analysis now.


    5.2.          Econometric analysis of data from Suceava

Our analysis of Suceava relies on variations of Models 1 and 2 in Baldwin et al. (1997).

        5.2.1. Data

A wealth of information was collected for Suceava by the IRIS team in Romania, including

auction, firm, and tract information. Auction data consists of the actual order and amount of bids

for each tract. Firm information is mostly location, plus, for four 2002 auctions (March 5, April


3370.2 and 61.4, respectively.



                                                   33

20, April 30, and June 19), contract information as well as the maximum authorized annual

processing volume of timber. Tract data includes tract location information (such as forestry,

production, and management units), volume, area, reserve price, bidding step, various forestry,

cutting, and product codes, as well as the high bid and the number of bids.

        Table 6 summarizes the number of registered bidders for each tract (by auction date) for

tracts sold separately in the 27 different auctions for which we have data. We observe that

"large" auctions tend to be followed after a few days by much smaller ones (presumably

secondary auctions for tracts which failed to sell at the primary auctions) which appear to be

much less competitive, assuming that the mean number of registered bidders per tract gives an

indication of the degree of competition to buy the tracts for sale. Over time, it appears that the

mean number of bidders has been increasing: the three largest means for the number of bidders

are to be found among the four 2002 auctions. It does not imply, however, that auctions have

become more competitive, as shown by the mean number of bids per tract in Table 7. With the

exception of the March 5, 2002 auction, the big auctions of 2002 (in April and June) rank second

and fourth from the bottom in terms of the mean number of bids per tract. Worse, there is on

average only one bid per registered bidder for these two auctions.

        For this report, we focus on the April 20, 2002 and June 19, 2002 auctions. The main

reason for this choice is the availability of contract information, which also tells us about supply

constraints for the firms participating in these auctions. This information is critical for our

methodology, recognizing that what could be interpreted as the result of collusive activities may

in fact be due to changes in the supply of timber. After eliminating records for non-mature

timber and for tracts sold in groups or tracts with only one registered bidder, as well as

incomplete or clearly erroneous records (e.g., a few tracts had a negative area) and tracts with

less than 100 cubic meters of timber (an unusually small amount), we are left with 615

observations: 387 for the April 20, 2002 auction and 228 for the June 19, 2002 auction.

        Table 8 presents a summary of some of the key variables for the tracts we consider.

Unfortunately, we have little direct information about the cost of harvesting timber for each tract.

Although location may play a role (see Table 9 for forest divisions in Suceava), we need to rely

on variables such as area (ceteris paribus, it may be costlier to harvest timber on a larger area), or




                                                   34

cod_produs (see Table 10). Unfortunately, there is no variation of harvesting system in our

sample (see Table 11).

       Ideally, following Baldwin et al. (1996), we would like to have information about the

inventory of the auction participants. A higher inventory would likely lessen the urgency to

acquire new timber and therefore should lead to lower timber prices. Since this information is

unavailable, we construct a crude index of the supply situation by dividing the volume of timber

under contract for the auction participants (at the beginning of each auction) by their annual

authorized timber volume. This information is stored in the variable CapRatio in Table 8.

       For the 615 tracts of mature timber we analyze, the volume of timber ranged from 100 to
11420 cubic meters (mean: 1521 m3); tract area varied between 0.05 ha and 512.7 ha, with a
mean of 71.5 ha. The corresponding high bids per m3 ranged from a low of 173,000 lei to a high

of 1,009,000 lei, with a mean of 480,944 lei (1 USD = 33,933.50 lei). The mean number of bids

per tract was 8.54, with a minimum of 1 and a maximum of 46.

       5.2.2. Theoretical models and econometric approach

Throughout this analysis, we assume that the hypotheses of the standard independent private

values (IPV) model hold (see Milgrom and Weber 1982). The IPV framework seems to make

sense here because mills often have different production costs and holdings of uncut timber. In

addition, a number of theoretical results are available in this framework and not in others.

       Key assumptions of the IPV model include:

   · Bidders are risk neutral.

   · All bidders are equally well informed about the quality and quantity of timber on any

       given tract. This is one reason why we are restricting our analysis to mature timber sales,

       which are almost all second-growth; second-growth timber is typically more homogenous

       than old-growth timber.

   · Bidders also have the same information about future potential price movements, although

       in practice large firms may have an information advantage if they are involved in export

       markets or have better government connections.

Although some of these assumptions are not very satisfactory, they are necessary to implement

our theoretical models.



                                                   35

        As a starting point, we consider two models. One of their main roles is to explain low

winning bids. One potential reason is the presence of collusion, but this is by no means the only

possible explanation. Demand for timber could also be depressed by a number of factors, which

we try to capture through covariates. These factors could be linked for example to timber quality

or to supply constraints. We try to capture some of the latter through the variable CapRatio.

        As in the analysis of Neamt auctions, we define:

        Bi     Winning bid for tract i (lei);

        ri     Reserve price for tract i (lei);

        vi     Volume of timber in tract i (cubic meters).

In addition, we define

        ni     Number of registered bidders for tract i.

        We also suppose that observed winning bid per cubic meter, ln(Bi/vi) is normally
distributed with mean zi and variance 2 where zi is a vector of covariates. This is a convenient

assumption also used in Baldwin et al. (1997). Covariates considered include:

    · Volume of timber on a tract for sale (cubic meters). Indeed, it has been suggested that

        there are increasing returns with volume of a tract.

    · Square of the volume of timber, for non-linear effects.

    · Area of a tract, in hectares. We supposed that the area of a tract could impact logging

        costs.

    · Dummy variables for the NFA ID codes for forest divisions, to capture unobserved

        specific circumstances in different forest divisions (see Table 9).

    · Type of timber sale (cod_produs; see Table 10).

    · CapRatio, which equals the volume of firms partaking in an auction divided by their

        annual authorized volume of timber.

        Unfortunately, we had no data on actual logging costs or timber quality. A first step of

our model selection process is to select what covariates we want to use in our two models.

Preliminary regression helped us make a first selection among these covariates. We found that

dummy variables for forest divisions were all insignificant, so we did not include them in the

remainder of this analysis. Likewise, timber type (cod_produs) was not significant at 5%. By



                                                   36

contrast, all the other covariates are strongly significant and have the expected signs: positive for
volume (more timber increases the value of a tract), negative for volume2 (there may be

diminishing scale economies with the volume harvested, as for Neamt), negative for area (less

timber per unit area decreases value), and strongly negative for CapRatio (as firms are close to

their annual maximum harvest, they have less of an incentive to purchase rights to new timber).

Table 12 presents a regression without the forest division variables.

       As mentioned in the previous section, we observe only the number of bidders willing to

pay at least the reserve price. We should therefore model the value distribution as a distribution

censored from below by the reservation price, which varies from tract to tract. In our dataset of

615 observations, 94 observations are within 1.055 of the reservation price. In order to take

advantage of the information contained in these observations, we censor them by discarding their

actual value but we retain that they are within 5.5% of the reserve price.

       We are now ready to present our 2 models. Model 1 assumes non-cooperative behavior in

an English auction with unit supply. It follows that the winner will pay the second highest

valuation overall. Baldwin et al. (1997) show that the corresponding likelihood function is

                                       
                           L(,;D) = iIa  bi    vi                                       
                                                    f2(ui | ri,ni ) F2(yi | ri,ni ),
                                                                                                (3)
                                                                   iIb

where:

        is the unknown vector of coefficients of the covariates zi;
       2 is the unknown variance of the distribution of ln(Bi/vi);

       D is our data set;

       Ia is the set of uncensored high bids;

       Ib is the set of censored high bids;

        ui =  ln(bi / vi ) -  'zi      ln(ri / vi ) -  'zi         ln(1.07 ri / vi ) -  'zi.
                                 , si =                    , yi =            

       ni is the numbers of bidders for auction i;

       f2(ui|ri,ni) and F2(ui|ri,ni) are respectively the density and the distributions of the 2nd order

       statistics of n independent draws from a standardized normal distribution truncated from

       below by ri.



                                                        37

         More generally, if f(ui|ri,ni) and F (ui|ri,ni) are respectively the density and the

distributions of the th order statistics of n independent draws, from a standardized normal

distribution truncated from below by ri, then


                 f (u | r,n) =          n!
                                  (n -)!( -1)!1-(u)  (u) -(s) 
                                                     (u)  1-(u)  1(u-( (s)            ) -    n
                                                                                               , (4)
                                                                                         s)

and
                                         -1
                          F (u | r,n) =   (      n!
                                                                                           ,       (5)
                                          j=0 n - j)! j! (u) -(s) 
                                                           1-(u) j  1(u-(          ) -(s) n

                                                                                      s)

where (·) and (·) are respectively the standard normal density and cumulative distribution

functions.

         Model 2 is a model of collusion in an English auction with unit supply. If a coalition
contains k bidders with the k-highest values and not the k+1st one, then the coalition will pay the

latter. Using the same notation as for Model 1, the likelihood function for Model 2 is

                                            
                          L(, ,;D) = iIa     bi   vi                                     
                                                      hc (ui | ri,ni ) Hc (yi | ri,ni ),
                                                                                                   (6)
                                                                      iIb

where:

                                    n                                                n
                 hc (ui | ri,ni ) =qi   c  f (ui | ri,ni ), Hc (yi | ri,ni ) = pi +
                                                                                c  qiF( c  yi | ri,ni ), (7)
                                   =2                                              =2

and qi =1- pi , qi = pi
        c        2    c       -1
        2                          (1- pi ) for =3 ni. In addition, the probability that a bidder

joins a coalition at auction i is given by


                                    pi =  e
                                        1+ e   .                                                   (8)

In the above,  is an unknown parameter that needs to be estimated from the data.

         Models 1 and 2 were estimated using a program written in GAUSS after managing the

data in STATA. It should be emphasized that the log likelihood functions for Models 1 and 2 are

difficult to estimate because they admit local maxima. After unsuccessfully trying to find their

maximum using the Gauss procedure optimum, we switched to a simulated annealing algorithm




                                                         38

which is much more robust and much more reliable for maximizing functions of several

variables (see Corana et al., 1987).

        5.2.3. Results

Results of our maximum likelihood estimations are summarized in Tables 13A and 13B. Our

best estimates for both models do not include volume squared, which no longer appears to be

significant. For Model 1, we see that all variables are significant (at 5% at least) and they have

the expected sign (see the discussion above). For Model 2, we obtain very interesting results:

volume is now insignificant, and most importantly pi, the estimate of the probability that a bidder

joins a coalition at auction i is now quite high and very significant: inserting =1.2358 from

Table 13B in (8) leads to pi=0.775. The signs of the other coefficients are as expected. In

addition, we find that the value of the logarithm of the likelihood function for Model 2 is much

higher than for Model 1; a likelihood ratio test is significant at 1%. Note, however, that caution is

necessary when interpreting this test because the probability of collusion is at the boundary of its

permissible domain under the null hypothesis; we know from Andrews (2001) for example, that

this situation causes difficulties similar to unidentification so standard asymptotics may fail. An

approach similar to the one used by Khalaf, Saphores, and Bilodeau (2003) in a different context

may be used to address this problem.

        These results suggest nevertheless that the collusive model with unit supply (Model 2)

clearly outperforms our non-cooperative model with unit supply (Model 1) for the auctions that

took place in Suceava on April 20, 2002 and on June 19, 2002.

        5.2.4. Conclusions: Suceava

These results suggest the presence of collusion during the 2002 forestry auctions in Suceava.

They are in line with anecdotal comments gathered by the IRIS field research team during the

collection of the data used in this report. It is important to emphasize that these findings are not

very surprising, given results from studies that found evidence of collusion in forestry auctions in

the United States for example (see Baldwin et al., 1996).

        However, these results are still preliminary. In future work, the lognormal assumption

should be checked; more elaborate auction models, especially Models 3-5 in Baldwin et al.




                                                    39

(1987), should be evaluated; and a better parameterization of the probability that a firm will join

a coalition should be attempted.



6.0     Conclusions

This paper has attempted to detect collusion in Romanian forestry auctions for two forest

directorates, Neamt and Suceava, using two different econometric methodologies to reflect data

availability. Neamt was the first forest directorate in Romania to introduce timber auctions.

Unfortunately, relatively little data was available electronically. For example, the research team

could not get electronic records of important tract characteristics such as the composition of

timber by size class and species group, which contributes to the value of a tract, or the distance

of a tract to the nearest forestry road, which contributes to harvesting costs. A lack of resources

also precluded the collection of information about mill location and the density of forestry roads.

In addition, detailed bidding information including the number of bidders for each tract was

unavailable. Some of these limitations also existed for Suceava.

        The data analyzed for Neamt is a subset of scheduled sales of mature timber between

1999 and 2002. Our testing procedure contrasted the maximum likelihood of a simple regression

model in which winning bids were assumed to equal bidders' valuations for all tracts, to a

censored regression model that isolated tracts where the winning bid was within 7% of the

reservation price. We found that the censored regression model fits the data better than the

standard linear regression model, so we cannot rule out the presence of collusion, although

collusion is not the only possible explanation for this better fit.

        Our analysis of timber auctions in Suceava, which focused on 2002 auctions, provides

stronger evidence of collusion. Detailed bidding information, including the order and the size of

each bid, was available, and this allowed us to use more powerful econometric models that also

start to account for supply effects. In our analysis, we contrasted two models in an English

auction with unit supply: Model 1 assumed competitive behavior, while Model 2 assumed that a

subset of bidders colludes to lower the winning bid. Our results show that the collusive model

(Model 2) strongly outperforms the competitive model (Model 1) for the auctions analyzed.

        While these models are not yet entirely satisfactory--partly because auction theory is still

lagging behind actual auction design and practice but also because of potential econometric

complications--these results suggest the presence of collusion in Romanian forestry auctions.


                                                   40

These findings reinforce anecdotal evidence gathered by the IRIS field team in Neamt, where

just three firms harvest more than half of the total amount of timber sold in that directorate and

are alleged to engage in collusive bidding.

        Collusion in auctions is by no means unique to Romania or to developing economies. As

discussed in section 4, collusion has also been detected in U.S. Forest Service auctions and in

other U.S. markets where auctions are used, including highway construction and milk contracts

for schools. Auctions offer a very useful mechanism for allocating contracts, however, and they

are likely to become increasingly popular as developing economies try to increase their

competitiveness. As in other markets, it is necessary to ensure transparency and the respect of

ground-rules to maintain competition and confidence in the economic institutions. In this regard,

it is reassuring that increasingly sophisticated and powerful econometric methods are becoming

available to detect instances where these rules are violated and collusion occurs.

        For Romania, this study suggests that efforts should be made to systematically collect and

make available data to enable ongoing analyses of forestry auctions. In addition to information

on individual bids for each tract (bid level, bid order, and bidder identity), these data should

include basic firm information (such as mill location and capabilities) and tract information (tree

species and their corresponding volume as well as basic parameters that determine harvesting

costs). Even with enhanced electronic availability of data, we doubt that "real-time" analysis of

Romanian timber auctions will be possible. Instead, we envision a situation in which analysts,

perhaps local academics working under contract for the Ministry or some other appropriate

government agency, would annually apply the sorts of methods presented in this report to data

from auctions during the past year.          They would prepare a sort of "Timber Auction

Competitiveness Report." The aim would be identify forest directorates where collusion is more

likely to be occurring. The government could then use this information to conduct targeted

investigations and implement measures to reduce collusion and its consequences.

        What types of anti-collusion measures are available to the government? Discussion of

this point in HIID (1996, p. 2) remains relevant:

        Governments can take several actions to reduce the risks and consequences of imperfect
        competition in stumpage auctions. One is to adopt and enforce anti-collusion laws.
        Reportedly, such a law will go into effect in Romania on January 1, 1997. A second is to
        minimize (but not necessarily eliminate) the requirements that bidders must satisfy to
        participate in auctions. ... A third action is to coordinate the dates of auctions with the



                                                  41

         forestry calendar, to ensure that more timber is offered for sale when demand (and thus
         potential competition) is high. Romsilva [i.e., the NFA] faces rules [related to the AAC
         allocation] that reportedly violate this principle. ... A fourth action ... is to employ
         starting prices (reserve prices) as a means of preventing eventual winning prices from
         being too far below fair market values.

Klemperer (2004), in a recent and influential book on auction design, emphasizes low entry

barriers and economically meaningful reserve prices as "what really matter" in fighting
collusion.34 The most obvious entry barrier in Romanian timber auctions is the prohibition

against foreign participation.          The government could consider relaxing this rule in forest

directorates where there is evidence of collusion among domestic companies. Publication of

forest management plans might also lead to a more rational spatial distribution of mills with

regard to available timber supplies, reducing the number of instances where only a few mills are

located in close proximity to areas where many tracts are scheduled for auction.

         The HIID (1996) report was motivated by the government's interest in developing

improved procedures for setting reserve prices, in part to combat collusion. It concluded that the

government could make better use of winning prices in past auctions to set reserve prices in

upcoming auctions. This method is termed transactions analysis. It amounts to setting the

reserve prices for new tracts equal to winning prices for comparable tracts that sold in past

auctions, minus a margin to reduce the risk of overstating the value of a tract and causing it to go

unsold.     Statistical techniques are used to calculate comparable prices. This method is used

routinely in timber auctions in western countries. A disadvantage is that if collusion is common,

then basing reserve prices on winning prices from past auctions will simply perpetuate the

downward pressure of collusion on winning prices.                   A solution to this problem is to use

techniques like the ones presented in this paper to identify tracts where bidding appears to have

been affected by collusion and remove them from the database used to set reserve prices.

         A second method is residual value analysis, which considers the difference between log

prices and logging costs. This method is also commonly used, but it relies on information that is


34A fine (but not unimportant) point is whether governments should reveal their reserve prices to bidders. Policies

vary in recently created timber auction systems. In Cameroon, as in Romania, reserve prices are made public before

auctions. In Ghana, however, they are revealed only if none of the received bids exceeds them. Revealing reserve

prices to bidders certainly makes it easier for colluding bidders to determine how low they can go and still be

awarded a tract.



                                                         42

less directly observable by the government (i.e., information about logging costs). Largely for

this reason, the HIID report concluded that transactions analysis was the more suitable method

for setting reserve prices in Romanian timber auctions. This conclusion is even more likely to be

correct today, given the decade of additional data on winning bids and tract characteristics that

can in principle be used to establish the market value of tracts yet to be auctioned. Implementing

transactions analysis will be hindered in Romania, however, until a more comprehensive and

consistent effort is made to computerize auction data across forest directorates.




                                                  43

          Table 1. Scheduled sales of mature timber at Romanian timber auctions held during January-July, 2003


                          All species       Coniferous           Beech              Oaks             Mixed            Mixed

               Forest                                                                             hardwoods        softwoods

   Region   directorate Volume   Winning  Volume  Winning    Volume Winning   Volume  Winning   Volume Winning   Volume  Winning
                         (m3)*      bid              bid                bid               bid             bid                bid
                                (lei/m3)** (m3)*  (lei/m3**) (m3)*  (lei/m3)** (m3)*  (lei/m3)** (m3)* (lei/m3**) (m3)*  (lei/m3)**
Central     Alba         96,154 597,356    44,171 615,862    39,730 602,580     5,076   658,134   6,610 410,646     567 422,227
            Brasov       58,961 542,455     8,540 660,718    44,720 541,285     1,797   444,857   3,904 342,088
            Cluj         58,600 575,598    19,745 746,163    29,447 538,809     1,985   699,235   7,423 234,780
            Hunedoara   180,195 580,241    47,481 600,101 120,996 585,281       6,806   531,170   4,912 332,106
            Salaj        28,391 974,933          2 281,000   16,210 1,044,64    9,021 1,035,507   2,629 425,968     529 536,554
            Sibiu        70,491 705,101    34,609 666,902    20,584 781,282     7,908   839,604   7,085 532,877     305 411,536
Danube      Constanta    30,990 466,872                                                           3,989 570,920 27,001 451,501
            Dolj         83,248 474,746                                        14,013   568,509 16,536 437,329 52,699 461,555
            Ialomita    185,739 511,557                                                           9,783 841,972 175,956 493,186
            Mehedinti   106,420 451,671                      77,656 458,008     9,227   524,224   5,986 366,136 13,551 403,738
            Olt          23,663 522,829                                         6,690   742,113   3,803 558,041 13,170 401,270
            Tulcea       47,587 448,046                                         1,217   438,000   3,444 340,000 42,926 457,000
East        Botosani     25,878 944,329                       11,8521,203,84    2,071 1,498,226   3,048 750,752    8,907 536,458
            Iasi         98,859 908,310        373 667,716   19,858 916,193 14,840 1,579,692 26,090 848,010 37,698 683,977
            Vaslui       42,914 681,868                       13,073 817,807    5,021 1,021,284   8,952 456,716 15,868 589,496
            Vrancea     150,768 892,515    10,467 695,421    94,349 968,199     7,801 1,954,600 12,489 654,621 25,662 487,556
Northeast   Bacau       361,430 638,880    76,876 529,606 263,705 667,666       2,334   786,530   3,675 565,255 14,840 688,440
            Covasna      55,490 661,540    20,537 701,115    32,527 603,174     2,426 1,109,069
            Harghita     84,185 680,674    77,590 697,738      6,595 479,912
            Maramures   231,376 663,994    68,345 591,492 153,813 672,022       8,110 1,151,127   1,108 456,085
            Mures       187,336 648,292 121,641 675,714      49,382 596,845     5,571   797,703   9,170 494,216    1,572 511,760
            Neamt       466,192 660,448 333,592 651,406 109,721 674,213         4,736 1,168,285 10,515 691,121     7,628 500,297
            Suceava     1,444,07 555,539  1,399,58 552,225   44,489 659,824                                            4 83,820




                                            44

Table 1. Scheduled sales of mature timber (continued)


                             All species         Coniferous            Beech                  Oaks               Mixed             Mixed

             Forest                                                                                            hardwoods        softwoods

 Region directorate Volume           Winning   Volume  Winning    Volume     Winning    Volume   Winning    Volume  Winning   Volume Winning
                           (m3)*        bid               bid                   bid                 bid                bid               bid
                                    (lei/m3)**  (m3)*  (lei/m3**)  (m3)*    (lei/m3)**   (m3)*  (lei/m3)**   (m3)*  (lei/m3**) (m3)* (lei/m3)**
Northwest/
West       Arad           171,584 716,942         2,248 415,651     71,716 777,235 60,497          783,504 28,759 432,769       8,364 776,610
           Bihor          122,060 631,882        12,697 755,056     73,243 603,479 20,341          722,976 14,950 542,037        829 639,937
           Satu Mare        35,753 886,517        1,822 540,000     13,575 851,223 17,407 1,004,050            2,456 523,280     493 798,645
           Timis          127,006 745,893         4,786 602,593     44,805 683,461 54,440          848,410 17,625 650,246       5,350 668,852
South      Arges          118,279 813,461        13,321 870,143     94,425 766,888        6,960 1,491,153      3,515 514,632       58 403,000
           Bistrita N.       4,084 462,271        4,024 464,000                              33    444,700       27 226,000
           Bucuresti        68,775 689,448                                               12,139 1,264,595 13,877 854,625 42,759 472,561
           Buzau          112,553 653,925        32,278 543,242     50,334 609,616 11,906 1,138,211 13,885 756,021              4,150 321,228
           Dāmbovita        59,720 915,377          805 680,000     24,236 637,469 27,381 1,229,930            4,749 744,517    2,549 571,504
           Gorj           130,915 571,003         5,218 528,095   113,784 568,504         7,827    617,442                      4,086 606,425
           ICAS           217,376 800,273       155,952 832,307     53,695 663,738        5,637 1,352,552      2,080 426,968       12 705,196
           Prahova        154,280 649,811        45,892 767,423     92,835 581,318        6,535 1,121,260       900 298,554     8,118 427,618
           Teleorman        26,594 613,040                                                9,467    994,821      940 625,000 16,187 389,060
           Valcea         152,901 692,214        16,926 745,679   104,145 642,104 20,542 1,049,892             6,720 357,901    4,568 519,918
Southwest  Caras Severin  323,620 630,040       17,533 532,432 233,782 633,176 27,289 777,975 21,139 574,906 23,877 550,743

  Totals                 5,944,445 639,154 2,577,056 607,050 2,119,282 661,918 405,051             950,320 282,773 579,667 560,283 505,779

Source: Romanian National Forestry Utility
Notes:  *Volumes are totals for the 7-month period.
       **Winning bids are weighted-average means. Winning bids in italics are ones that are below the national mean.




                                                   45

        Table 2. Summary data on timber sold by auction in Neamt Forest Directorate


      Characteristic             Type of sale                        Year
                                                        1999   2000   2001   2002  2003*
Total volume             Scheduled: mature timber        263.6  476.3  477.9  157.0 100.3
(thousand m3)            Scheduled: thinning              73.8   74.4  119.4   69.2  93.1
                         Salvage: mature timber          140.3   86.1   64.2  459.9  63.6
                         Salvage: immature timber          7.0    0.8    2.7    9.9   2.7
                         Salvage: diseased timber          0.8    6.9   22.3    1.6   0.0
                         Total                           485.5  644.5  686.7  697.7 259.7
                         % Scheduled                      69%    85%    87%    32%   74%

Total revenue            Scheduled: mature timber         85.6  280.8  272.7   93.1  72.7
(billion lei)            Scheduled: thinning               8.2   14.6   26.5   13.9  29.6
                         Salvage: mature timber           34.1   20.9   31.4  242.5  38.0
                         Salvage: immature timber          1.8    0.1    1.2    2.5   1.1
                         Salvage: diseased timber          0.1    2.6    6.3    0.5   0.0
                         Total                           129.8  319.0  338.2  352.4 141.3
                         % Scheduled                      72%    93%    89%    30%   72%

Mean winning bid         Scheduled: mature timber         325    590    571    593    724
(thousand lei per m3)    Scheduled: thinning              110    196    222    201    318
                         Salvage: mature timber           243    243    489    527    597
                         Salvage: immature timber         261    120    420    249    402
                         Salvage: diseased timber         119    369    283    313      -

Source: Romanian National Forestry Utility, Neamt branch.
* For 2003, January-June only.




                                                   46

      Table 3. Summary data on timber sold by negotiation in Neamt Forest Directorate


      Characteristic            Type of sale                       Year
                                                         1999   2000   2001   2002
Total volume            Scheduled: mature timber           22.3   48.0    0.9   29.1
(thousand m3)           Scheduled: thinning                39.4   40.5    6.5    0.0
                        Salvage: mature timber             49.8   39.2    2.5   14.5
                        Salvage: immature timber            3.7    1.7    0.1    1.3
                        Salvage: diseased timber            0.1    3.2    0.0    0.0
                        Total                            115.3  132.5     9.9   45.0
                        % Scheduled                       54%    67%    74%    65%

Total revenue           Scheduled: mature timber            3.7   11.7    0.4   12.7
(billion lei)           Scheduled: thinning                 3.0    3.2    0.7    0.0
                        Salvage: mature timber              5.9    6.8    0.6    7.1
                        Salvage: immature timber            0.3    0.2    0.0    0.3
                        Salvage: diseased timber            0.0    0.4    0.0    0.0
                        Total                              12.9   22.4    1.8   20.1
                        % Scheduled                       52%    67%    65%    63%

Mean price              Scheduled: mature timber           164    243    468    437
(thousand lei per m3)   Scheduled: thinning                 77     80    114       -
                        Salvage: mature timber             118    174    244    492
                        Salvage: immature timber            83    122    173    204
                        Salvage: diseased timber            75    137       -   205

Source: Romanian National Forestry Utility, Neamt branch.




                                                   47

                 Table 4. Estimation results for model (1), Neamt.

       Generalized linear model                            No. of obs   =    869
       Optimization      : ML: Newton-Raphson              Residual df  =     809
       Log likelihood = -122.9747963                       AIC        = .4211158
       _____________________________________________________


         ln(Bi/vi) |   Coef.     Std. Err.  z       P>|z|   [95% Conf. Interval]
       _____________________________________________________


         ln(ri/vi) | 1.06333     .0491702 21.63 0.000       .9669586 1.159702
              vi  | .0000641     .0000232    2.77 0.006     .0000187 .0001095
             vi2   | -9.79e-09   4.08e-09 -2.40     0.016 -1.78e-08 -1.80e-09
        Constant | -.2070782     .5604832 -0.37 0.712 -1.305605 .8914486
       _____________________________________________________


Note: Model also includes 23 dummy variables for forest division (dummy for Bicaz is omitted)
and 34 dummy variables for auction date (dummy for May 14, 1998 auction is omitted).



                 Table 5. Estimation results for model (2), Neamt.


       Censored normal regression                 Number of obs    =     869
                                                  LR chi2(59)      =   918.92
                                                  Prob > chi2      =   0.0000
       Log likelihood = -60.849009                Pseudo R2        =   0.8831
       ___________________________________________________________

         ln(Bi/vi) |   Coef.     Std. Err.  t              P>|t|  [95% Conf. Interval]
       ___________________________________________________________

         ln(ri/vi) | .9819834    .0551903 17.79            0.000   .8736505 1.090316
              vi  | .0000733     .0000269    2.73          0.007   .0000205    .000126
             vi2   | -1.04e-08   4.66e-09 -2.24            0.025 -1.96e-08 -1.29e-09
        Constant | .7978003      .6301034    1.27          0.206 -.4390278 2.034628
       ___________________________________________________________

            _se | .2514063       0079685        (Ancillary parameter)
       ___________________________________________________________

        Obs. summary:            433 uncensored observations
                                 436 right-censored observations

Note: Model also includes 23 dummy variables for forest division (dummy for Bicaz is omitted)
and 34 dummy variables for auction date (dummy for May 14, 1998 auction is omitted).




                                              48

        Table 6. Statistics on the number of registered bidders for tract by auction date

         ___________________________________________________________

         Auction          # of                             Standard        Min      Max
         Date             Tracts          Mean             Deviation
         _______________________________________________________________________

         09 Feb 00        230             3.31             1.92            1        10
         16 Feb 00        34              1.26             0.62            1        3
         15 Mar 00        75              2.12             1.05            1        5
         22 Mar 00        28              1.25             0.52            1        3
         07 Apr 00        147             3.01             1.76            1        8
         21 Apr 00        41              1.32             0.65            1        3
         14 Jun 00        106             2.21             1.04            1        6
         16 Jun 00        25              1.16             0.37            1        2
         25 Aug 00        156             3.62             1.77            1        10
         30 Aug 00        17              1.53             0.51            1        2
         08 Sep 00        170             5.28             2.64            1        12
         13 Sep 00        16              1.38             0.72            1        3
         06 Dec 00        83              6.01             2.53            1        12
         09 Feb 01        283             4.88             2.65            1        12
         14 Feb 01        22              1.14             0.35            1        2
         02 May 01        364             6.52             3.46            1        18
         09 May 01        19              2.00             0.75            1        3
         29 Jun 01        355             7.62             3.64            1        18
         04 Jul 01        13              3.00             0.41            2        4
         31 Aug 01        284             4.80             2.35            1        12
         04 Sep 01        17              2.24             0.90            1        4
         31 Oct 01        417             6.89             3.21            1        21
         07 Nov 01        12              2.42             0.67            2        4
         05 Mar 02        228             8.73             4.12            1        20
         20 Apr 02        450             8.48             4.06            1        28
         30 Apr 02        6               2.50             1.38            1        4
         19 Jun 02        371             9.76             5.30            1        26
         All              3969            6.06             4.07            1        28
         _________________________________________________________________________

Notes. These statistics are based on the "Ofert" file, which includes only tracts sold separately. For 397
tracts, there was a single registered bidder.




                                                       49

                     Table 7. Statistics on the number of bids by auction date

       ___________________________________________________________

       Auction           # of                            Standard      Min        Max
       Date              Tracts           Mean           Deviation
       _______________________________________________________________________

       20000209          190              8.34           10.19         1          49
       20000216          8                2.13           2.48          1          8
       20000315          56               10.34          12.97         2          51
       20000322          6                2.33           816           2          4
       20000407          110              13.24          13.22         1          54
       20000421          9                8.89           9.56          1          29
       20000614          83               5.70           12.90         1          100
       20000616          4                1              0             1          1
       20000825          140              8.91           10.67         1          47
       20000830          7                3.71           7.18          1          20
       20000908          158              10.70          12.03         1          73
       20000913          4                2.25           0.96          1          3
       20001206          81               11.99          10.20         1          37
       20010209          262              10.53          13.64         1          56
       20010214          6                5.33           9.67          1          25
       20010502          342              13.20          17.68         1          95
       20010509          20               1              0             1          1
       20010629          348              14.36          17.46         1          75
       20010704          16               1              0             1          1
       20010831          263              9.38           12.01         1          73
       20010904          15               2.53           1.19          2          6
       20011031          409              15.89          14.32         1          69
       20011107          12               2.58           1.00          2          5
       20020305          231              22.22          16.80         1          102
       20020420          641              8.87           9.38          1          70
       20020430          8                3.63           4.31          1          13
       20020619          790              8.04           7.48          1          48
       20020625          58               3.26           3.88          2          19
       All               4279             10.96          13.07         1          102
       ________________________________________________________________________

Notes. These statistics are based on the "auction_data_2000-2001-2002" file. It includes all tracts.




                                                      50

                   Table 8. Summary Statistics for the Auction Data Analyzed

                Auction Date: April 20, 2002.
                _____________________________________________________________________________________________________

                Variable                Obs        Mean                     Std. Dev.               Min                     Max
                ____________________________________________________________________________________________________________________________
                Volume                  387         1470.106               1233.288                108                     8678
                Area                    387         61.68269                65.85765                0.6                     512.7
                up                     387         3.45478                 2.00773                 0                       10
                cu                     387         9.18863                 6.496557                1                       21
                cod_produs              387         5.656331                0.4755471               5                       6
                cod_taiere              387         35                      0                       35                      35
                ResPrice               387         341257.5                41397.82                220221                  436790
                WinBid                 387         504423.8                 149951.1                232000                  1009000
                Nbids                   387         9.095607               8.064742                1                       46
                CapRatio               387         0.1892                  0                       0.1892                  0.1892
                ____________________________________________________________________________________________________________________________




                Auction Date: June 19, 2002.
                _____________________________________________________________________________________________________


                Variable                Obs        Mean                     Std. Dev.               Min                     Max
                ____________________________________________________________________________________________________________________________
                Volume                  228         1607.307               1415.558                100                     11420
                Area                    228         88.09772                84.61378                0.05                    423.2
                up                      228         3.175439                1.566419               0                        8
                cu                     228          14.59211               4.258739                 8                       21
                cod_produs              228         5.671053                0.4708646               5                       6
                cod_taiere              228         35                      0                       35                      35
                ResPrice               228         313829.7                47660.49                156107                  415518
                WinBid                 228         441092.1                 124478.6                173000                  809000
                Nbids                   228         7.600877               6.529054                1                       32
                CapRatio               228         0.2625                  0                       0.2625                  0.2625
                __________________________________________________________________________________________________________




Notes. Volume is in cubic meters. Area is in ha. Cod_produs 5 or 6 indicate felling in large areas affected
by windthrows. Cod_taiere is the NFA ID code for a harvesting system; "cu" is the NFA ID code for a
forest division (ocolul silvic); "up" is the production unit where tracts are located. ResPrice is the
reservation price in lei / m3. WinBid is the winning bid in lei / m3. Nbids is the number of bids per tract.
CapRatio is the ratio of timber under contract divided by the annual authorized amount of timber for all
firms who registered in an auction.




                                                                      51

           Table 9. NFA ID codes and names for forest divisions (ocolul silvic, cu)

 Forestry Unit                                             Chief of Forestry
      Code             COD            Forestry Unit                Unit        Foresty Unit Sub-units
        1          BROSTENI        OS Brosteni           ing.Palade Radu       Ocolul Silvic Brosteni
        2          CRUCEA          OS Crucea             ing.Palade Petru      Ocolul Silvic Crucea
        3          DORNACAN OS Dorna Candreni            ing.Boghean Petru     Ocolul Silvic Dorna Candreni
        4          COSNA           OS Cosna              ing.Birladeanu B.     Ocolul Silvic Cosna
        5          VATRADOR OS Vatra Dornei              ing.Carcea Maria      Ocolul Silvic Vatra Dornei
        6          IACOBENI        OS Iacobeni           ing.Fodor Mihai       Ocolul Silvic Iacobeni
        7          CIRLIBAB        OS Crlibaba           ing.Turcanu C.        Ocolul Silvic Crlibaba
        8          BREAZA          OS Breaza             ing.Coman Constantin  Ocolul Silvic Breaza
        9          POJORITA        OS Pojorta            ing.Solovastru V.     Ocolul Silvic Pojorta
       10          VAMA            OS Vama               ing.Malis Vasile      Ocolul Silvic Vama
       11          MOLDOVIT OS Moldovita                 ing.Selaru Vasile     Ocolul Silvic Moldovita
       12          FRASIN          OS Frasin             ing.Gingolia Dan      Ocolul Silvic Frasin
       13          STULPICA        OS Stulpicani         ing.Miros Radu        Ocolul Silvic Stulpicani
       15          SOLCA           OS Solca              ing.Cotoar Petru      Ocolul Silvic Solca
       16          MARGINEA OS Marginea                  ing.Totoescu Adrian   Ocolul Silvic Marginea
       17          PUTNA           OS Putna              ing.Chidovet Stelian  Ocolul Silvic Putna
       18          FALCAU          OS Falcu              ing.Mutescu Ioan      Ocolul Silvic Falcu
       19          BRODINA         OS Brodina            ing.Hlodec Vasile     Ocolul Silvic Brodina
       20          MALIN           OS Mlini              ing.Vod Petru         Ocolul Silvic Mlini
       21          RISCA           OS Rsca               ing.Stefan Dan        Ocolul Silvic Rsca
       22          FALTICEN        OS Flticeni           ing.Negura Gheorghe   Ocolul Silvic Flticeni


Note. "Cu" is the NFA ID code for a forest division (ocolul silvic).



                     Table 10. Types of Timber Sales (Cod_produs)

   Code                                                  Translation
      1          Salvage I (not in areas affected by windthrows I)
      2          Salvage I ((not in areas affected by windthrows II)
      3          Felling in areas affected by windthrows (I)
      4          Felling in areas affected by windthrows (II)
      5          Felling in large areas affected by windthrows (I)
      6          Felling in large areas affected by windthrows (II)
      7          Thinning
      8          Sanitary Felling




                                                         52

              Table 11. NFA nomenclature for harvesting systems (Cod_taiere)

Cutting              Cutting           Cutting                             Translation
 Code                  Type            Group                             Cutting Group
   1     rase pe parchete mici            1     Clearcuts
   2     rase in benzi                    1     Clearcuts in stripes
   3     succesive in margine de masiv    2     Uniform shelterwood system felling - stand edge
   4     succesive 1                      2     Uniform shelterwood system felling
   5     succesive 2                      2     Uniform shelterwood system felling
   6     succesive 3                      2     Uniform shelterwood system felling
   7     succesive 4                      2     Uniform shelterwood system felling
   8     succesive definitive             2     Uniform shelterwood system felling (final felling)
         progresive in margine de               Group/non-uniform shelterwood system felling (Progressive) -
   9     masiv                            3     stand edge
                                                Group/non-uniform shelterwood system felling (Progressive) -
   10    progresive 1                     3     final felling
   11    progresive 2                     3     Group/non-uniform shelterwood system felling (Progressive)
   12    progresive 3                     3     Group/non-uniform shelterwood system felling (Progressive)
   13    progresive 4                     3     Group/non-uniform shelterwood system felling (Progressive)
   14    progresive -racordare            3     Group/non-uniform shelterwood system felling (Progressive)
   15    cvasigradinarite 1               4     Group selection system felling
   16    cvasigradinarite 2               4     Group selection system felling
   17    cvasigradinarite 3               4     Group selection system felling
   18    cvasigradinarite 4               4     Group selection system felling
   19    cvasigradinarite 5               4     Group selection system felling
   20    cvasigradinarite 6               4     Group selection system felling
   21    cvasigradinarite definitive      4     Group selection system felling - final felling
   22    gradinarite                      4     Selection system and conversion to selection system felling
   23    transformare spre gradinarit     4     Selection system and conversion to selection system felling
   24    in cadrul lucr.de conservare     5     Conservation felling
   25    de cring simplu                  6     Coppice system felling
   26    de cring -cazanire               6     Coppice system felling including removal of stump
   27    de cring -scaun                  6     Coppice system felling (1-1.5m above ground)
   28    de refacere                      7     Clearcuts for forest type rehabilitation
   29    de substituire                   7     Clearcuts for forest type substitution
   30    de ingrijire                     8     Thinning
   31    igienizarea padurii              9     Sanitary felling
   32    t. de transformare in pasuni    10     Clearcuts in afforested pastures
   33    t. de rarire in pasuni impad.   10     Thinning in afforested pastures
   34    t. in aliniamente               11     Clearcuts of tree alignements
   35    t. de accidentale I             12     Salvage
   36    t. de accidentale II            12     Salvage

Notes. Cod_taiere is the NFA ID code for a harvesting system.




                                                    53

   Table 12. Preliminary Regression for Auctions held on April 20, 2002 & June 19, 2002

_____________________________________________________________________________________________
Data points * Uncensored observations: 521 * Censored observations: 94.
Valid cases:     615
Missing cases:   0                Deletion method:         None
Total SS:        56.091           Degrees of freedom:      609
R-squared:       0.142            Rbar-squared:            0.135
Residual SS:     48.102           Std error of est:        0.281
F(5,609):        20.228           Probability of F:        0.000

                                  Standard                          Prob           Stand.          Cor
Variable         Estimate         Error            t-value          >|t|           Estimate        with Y
Constant         13.306           0.0733           181.49           0.000          ---             ---
Volume           0.128            0.0187           6.86             0.000          0.5526            0.21
Volume^2         -0.0123          0.0028           -4.46            0.000          -0.3550         0.11
Area             -0.0625          0.0158           -3.96            0.000          -0.1539         -0.14
Timber Type      -0.0200          0.0243           -0.82            0.411          -0.0313         -0.08
Cap. Ratio       -1.668           0.3253           -5.13            0.000          -0.1956         -0.21
_____________________________________________________________________________________________

Notes. The dependent variable is ln(Winning Bid per m3) if the winning bid is more than 1.055 times the
reservation price per m3; otherwise, it is censored and replaced by 1.055 * reservation price per m3.
Volume is in m3 /1000. Volume2 is in (m3 /1000)2. Area is in ha / 100. Only tracts with 100 cubic meters
or more of timber are analyzed here. Timber Type is a dummy variable for Cod_produs = 5. CapRatio is
the ratio of timber under contract divided by the annual authorized amount of timber for all firms who
registered in an auction. Estimates above are used as starting values for estimating Models 1 and 2.




                                                        54

                 Table 13A: Model 1: Parameters, Standard Errors, and MLE

                ______________________________________________________
                Parameter                  Value                  St. error
                ______________________________________________________
                Sigma                      0.696715               0.025895
                Constant                   11.595983              0.430996
                Volume                     0.011545               0.006752
                Area                       -0.044298              0.009011
                Cap. Ratio                 -5.636284              1.146541
                ______________________________________________________
                Log Maximum likelihood = -7590.72




                  Table 13B: Model 2: Parameters, Standard Errors, and MLE

                ______________________________________________________
                Parameter                  Value                  St. error
                ______________________________________________________
                Sigma                      0.658505               0.085333
                Constant                   13.943434              0.280554
                Volume                     -0.000753              0.020999
                Area                       -0.162008              0.080215
                Cap. Ratio                 -9.784077              2.307241
                Gamma                      1.235801               0.060740
                ______________________________________________________
                Log Maximum likelihood = -6901.56


Notes. The dependent variable is ln(Winning Bid per m3) if the winning bid is more than 1.055 times the
reservation price per m3; otherwise, it is censored and replaced by 1.055 * reservation price per m3.
Volume is in m3 /1000. Volume2 is in (m3 /1000)2. Area is in ha / 100. Only tracts with 100 cubic meters
or more of timber are analyzed here. CapRatio is the ratio of timber under contract divided by the annual
authorized amount of timber for all firms who registered in an auction. The dependent is assumed to be
normally distributed and sigma is an estimate of its standard deviation.




                                                       55

                                         References

Andrews, D.W.K., 2001, "Testing when a parameter is on the boundary of the maintained
        hypothesis," Econometrica 69: 683-734

Baldwin, L.H., R.C. Marshall, and J.-F. Richard, 1997. "Bidder Collusion at Forest Service
        Timber Sales", Journal of Political Economy 105(4): 657-699

Bajari, P. and G. Summers, 2002. "Detecting Collusion in Procurement Auctions", Antitrust Law
        Journal 70: 143-170.

Bajari, P. and L. Ye, 2001. "Deciding between Competition and Collusion", Working Paper,
        Stanford University (version May 18).

Brannman, L.E., 1996. "Potential Competition and Possible Collusion in Forest Service Timber
        Auction", Economic Inquiry 34: 730-745.

Corana, A., M. Marchesi, C. Martini, And S. Ridella, 1987. "Minimizing Multimodal Functions
        of Continuous Variables with the `Simulated Annealing' Algorithm," ACM Transactions
        on Mathematical Software, 13 (3 ­ September): 262-280.

Graham, D.A. and R.C. Marshall, 1987. "Collusive Bidder Behavior at Single-Object Second-
        Price and English Auctions", Journal of Political Economy 95 (December): 1217-1239.

Hendricks, K. and R.H. Porter, 1988. "An Empirical Study of an Auction with Asymmetric
        Information," American Economic Review 78: 865-883.

Hendricks, K. and R.H. Porter, 1989. "Collusion in Auctions," Annales d'Economie et de
        Statistique 15/16 (July-December): 217-230.

Harvard Institute for International Development, 1996. "Final Report: Transactions Analysis of
        Stumpage Auctions in Suceava, Romania." Report prepared for Central and Eastern
        European Environmental Economics and Policy Project and submitted to USAID.

Johnson, R.N., 1979. "Oral Auction versus Sealed Bids: An Empirical Investigation", Natural
        Resources Journal 19 (April): 315-335.

Khalaf, L., J.-D. Saphores, and J.-F. Bilodeau, 2003. "Simulation-based exact jump tests in
        models with conditional heteroskedasticity," Journal of Economic Dynamics and Control
        28 (2003) 531-553.

Klemperer, P., 2004. Auctions: Theory and Practice. Princeton University Press, Princeton, New
        Jersey.

Mattey, Joe P., 1990. The Timber Bubble that Burst: Government Policy and the Bailout of 1984.
        Oxford University Press, New York.

Mead, W.J., 1967. "Natural Resource Disposal Policy ­ Oral Auction versus Sealed Bids",
        Natural Resources Journal 7 (April): 194-224.

Milgrom, P.R. and R.J. Weber, 1982. "A Theory of Auctions with Competitive Bidding."
        Econometrica 50 (September): 1089-1122.

Pesendorfer, M., 2000. "A Study of Collusion in First-Price Auctions", Review of Economic
        Studies 67: 381-411.



                                                 56

Porter, R.H. and J.D. Zona, 1993. "Detection of Bid Riggins in Procurement Auctions", Journal
        of Political Economy 101: 518-538.

Porter, R.H. and J.D. Zona, 1999. "Ohio School Milk Markets: An Analysis of Bidding", Rand
        Journal of Economics 30(2): 263-288.

Robinson, M.S. 1985. "Collusion and the Choice of Auction," Rand Journal of Economics 16(1):
        141-145.

Saphores, J.-D., L. Khalaf, and D. Pelletier, 2002, "On jumps and ARCH effects in natural
        resource prices ­ An application to Pacific Northwest stumpage prices," American
        Journal of Agricultural Economics 84 (2): 387-400.

U.S. House of Representatives, House Committee on Agriculture, 1977. Timber Sales Methods
        and Related Practices, Ninety-Fifth Congress (Washington, D.C.: U.S. Government
        Printing Office).

Vickrey, W., 1961."Counterspeculation, Auctions, and Competitive Sealed Tenders", Journal of
        Finance 16 (March): 8-37.

Vincent, J.R., 1996. "Natural Resource Market Evolution in a Transition Economy." Paper
        presented in Environmental Economics Seminar, Kennedy School of Government and
        Department of Economics, Harvard University.

Waggener, Thomas R., 1969. "Some Economic Implications of Sustained Yield as a Forest
        Regulation Model." University of Washington, Institute of Forest Resources,
        Contribution No. 6. Seattle.

Weber, R.J., 1983. "Multiple-Object Auctions." In Auctions, Bidding, and Contracting: Uses and
        Theory, edited by R. Engelbrecht-Wiggans, M. Shubik, and R.M. Stark (New York: New
        York University Press).




                                                57

             Appendix: List of Individuals Interviewed (June-August, 2003)

Date Organization, place       Person (name and/or position)
19.06 ICAS, Bucharest          Mr Gheorghe Parnuta, general director
                               Mr Ovidiu Badea and Mrs Simona Dragoi, forest
                               management studies department
25.06 Ministry of Agriculture, Mr Adrian Popa, chief, department of management of private
      Forests, Waters, and     forests
      Environment, Bucharest
25.06 (same)                   Mrs Carmen Enescu, jurist
1.07  (same)                   Mr Gheorghe Marinescu, chief of forest fund department
                               Mrs Maria Moise
1.07  NFA headquarters,        Mrs Simona Dabulean
      Bucharest                Mrs Celsea

7.07  ITRSC (forest            Mr Melinte, chief
      inspectorate), Suceava   Mr Radu Sbarna
                               Mr Mihai Mihetiu
                               Mr Sorin Popescu

8.07  NFA headquarters,        Mr Corduneanu, director, forest fund
      Bucharest
8.07  Department of External   Mr Gica Draganescu, director
      Trade, Bucharest
17.07 NFA Forest Directorate,  Mr Ioan Tatoi, director
      Ialomita                 Mr Ioan Lorin, economist
                               Mr Cozan, forest fund department
18.07 NFA Forest Directorate,  Mr Francisc Toth, director
      Timisoara                Mr Baban, chief, forest fund department
25.07 NFA Forest Directorate,  Mr Marius Burciu
      Suceava                  Mr Iulian Stasiuc
27.07 Chamber of Trade and     Vice-president
      Industry
21.08 Petroforest,             Mr Teodorescu, director
      Piatra Neamt
21.08 Romanel,                 Mr Vasile Pupaza, chief of production department
      Piatra Neamt
21.08 New Star,                Mr Viorel Mestecaneanu, executive manager
      Piatra Neamt




                                             58