UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Essays in environmental economics Miyamoto, Takuro 2011

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2011_fall_miyamoto_takuro.pdf.pdf [ 843.92kB ]
Metadata
JSON: 24-1.0072234.json
JSON-LD: 24-1.0072234-ld.json
RDF/XML (Pretty): 24-1.0072234-rdf.xml
RDF/JSON: 24-1.0072234-rdf.json
Turtle: 24-1.0072234-turtle.txt
N-Triples: 24-1.0072234-rdf-ntriples.txt
Original Record: 24-1.0072234-source.json
Full Text
24-1.0072234-fulltext.txt
Citation
24-1.0072234.ris

Full Text

ESSAYS IN ENVIRONMENTAL ECONOMICS by TakuroMiyamoto B.A., Waseda University, 2003 M.A., Keio University, 2005 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in The Fa
ulty of Graduate Studies (E
onomi
s) THE UNIVERSITY OF BRITISH COLUMBIA (Van
ouver) September, 2011 © Takuro Miyamoto, 2011 Abstra
t The obje
tive of this dissertation is to improve our understanding of environmental poli
ies, par- ti
ularly with respe
t to two emerging alternative approa
hes to regulation. They are alternatives to the 
ommand and 
ontrol approa
h, whi
h poli
ymakers have relied on heavily sin
e the early 1970s. One alternative is to introdu
e market-based poli
y instruments like emission taxes or trad- able permits. Another alternative is to rely on voluntary approa
hes to environmental prote
tion. This thesis will study these two alternatives. Our rst essay will fo
us on voluntary programs (VPs) that aim to redu
e emissions of pollutants. We try to explain theoreti
ally why governments im- plement these programs and to examine the property of the VP whi
h the regulator implements to maximize so
ial welfare. We show that if setting an ef
ient mandatory standard is politi
ally dif
ult, a regulator might implement the VP be
ause it 
an generate higher so
ial welfare than the mandatory standard. The abatement rate of the VP that generates the highest so
ial welfare 
osts parti
ipating rms the same amount as the mandatory standard would. The se
ond essay will empiri
ally examine the determinants of environmental management system 
erti
ations, espe- 
ially the ISO 14001 
erti
ation, whi
h is a popular environmental pra
ti
e, and their impa
ts on environmental performan
e. In parti
ular, we fo
us on intra-industry spillovers of ISO 14001 adoption and environmental performan
e. We apply estimation methods of spatial e
onometri
s to a Japanese fa
ility's dataset to deal with the spillovers. We nd intra-industry spillovers of emis- sions redu
tion into the air between similarly sized fa
ilities and of ISO 14001 adoption between similarly sized fa
ilities that emit into water. The third essay will 
ompare taxes and quotas, when an informed polluting industry inuen
es them by politi
al 
ontributions to a government. We show that private information 
an improve so
ial welfare under taxes but 
annot improve it under quotas. Private information also redu
es a 
omparative disadvantage of the taxes over the quotas when the government does not 
are about so
ial welfare very mu
h. ii Table of Contents Abstra
t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix A
knowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x 1 Obje
tive and Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.1 Obje
tive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 2 Why Regulators Adopt Voluntary Programs: A Theoreti
al Analysis of Voluntary Pollutant Redu
tion Programs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 2.1 Introdu
tion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 2.2 Some fa
ts on voluntary programs . . . . . . . . . . . . . . . . . . . . . . . . . . 8 2.3 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 2.3.1 The legislative subgame . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 2.3.2 The VP subgame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 2.4 Comparative stati
 analysis of an equilibrium where the most effe
tive VP is im- plemented . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 2.5 Con
lusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 iii 3 Intra-Industry Spillover Effe
ts of ISO 14001 Adoption and Environmental Perfor- man
e in Japan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 3.1 Introdu
tion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 3.2 Ba
kground and hypotheses on spillover effe
ts . . . . . . . . . . . . . . . . . . . 31 3.2.1 ISO 14001 in Japan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 3.2.2 Hypotheses on spillovers . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 3.3 Estimation strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 3.3.1 Estimation of environmental performan
e equation . . . . . . . . . . . . . 37 3.3.2 Estimation of the ISO 14001 adoption equation . . . . . . . . . . . . . . . 37 3.3.3 Weight matri
es . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 3.3.4 Model 
hoi
e . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 3.4 Data des
ription . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 3.4.1 Emission data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 3.4.2 ISO 14001 adoption data . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 3.4.3 Data on other variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 3.4.4 Des
riptive statisti
s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 3.5 Estimation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 3.5.1 ISO 14001 adoption and performan
e of fa
ilities that emit into water . . . 45 3.5.2 Performan
e and ISO 14001 adoption of fa
ilities that emit into air . . . . . 50 3.6 Con
lusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 4 Taxes versus Quotas in Lobbying by a Polluting Industry with Private Information on Abatement Costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 4.1 Introdu
tion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 4.2 Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 4.3 Complete information 
ase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 4.3.1 Equilibrium tax rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 4.3.2 Equilibrium quota . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 4.4 In
omplete information 
ase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 4.4.1 Tax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 iv 4.4.1.1 Separating equilibria . . . . . . . . . . . . . . . . . . . . . . . . 67 4.4.1.2 Pooling equilibria . . . . . . . . . . . . . . . . . . . . . . . . . 72 4.4.2 Quotas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 4.4.2.1 Separating equilibria . . . . . . . . . . . . . . . . . . . . . . . . 74 4.4.2.2 Pooling equilibria . . . . . . . . . . . . . . . . . . . . . . . . . 75 4.4.3 Differen
e in results and in
entives between tax and quota . . . . . . . . . 76 4.4.4 Renements of beliefs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 4.5 Numeri
al examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 4.6 Con
lusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 5 Con
lusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 5.1 Summary of 
ontributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 5.2 Dire
tions for further resear
h . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 Appendi
es . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 Appendix A: Appendix for Chapter 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 A.1 A set of abatement rate and the number of parti
ipating rms that generates the highest aggregate abatement if N V P is not an integer . . . . . . . . . . . . . 97 A.2 Complete proof of the statement that there is equilibrium only when less than (N V P  1) rms parti
ipate in the VP and the mandatory poli
y is imple- mented (the se
ond part of Proposition 2.2) . . . . . . . . . . . . . . . . . 98 A.3 Proof of Proposition 2.3, 2.5 and 2.6 . . . . . . . . . . . . . . . . . . . . . . . 99 A.4 Proof of Proposition 2.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 A.5 Case without free-riding in lobbying . . . . . . . . . . . . . . . . . . . . . . . 100 Appendix B: Appendix for Chapter 3: Estimation results of WM IV and V models with m= 3;5, and 10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 Appendix C: Appendix for Chapter 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 C.1 Cal
ulations t HIC and t LIC . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 v C.2 Contribution under pooling equilibria when (1+ 
)ē L < 
E[ē℄, t < 
ē L and (1+ 
)ē L  
E[ē℄ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 C.3 Contribution under pooling equilibria when (1+
)ē L < 
E[ē℄ and when q> ē L and (1+ 
)ē L  
E[ē℄ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 C.4 Proof of Proposition 4.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 C.5 Proof of Proposition 4.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 C.6 Proof of Proposition 4.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 vi List of Tables Table 2.1 Overview of 33/50 program . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Table 2.2 Payoff matri
es when some rms parti
ipate (left) and when all rms parti
- ipate (right). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 Table 3.1 Top 3 for ISO 14001 
erti
ates in 2000, 2003, and 2008 . . . . . . . . . . . 32 Table 3.2 Des
riptive statisti
s of water emitting fa
ilities in 2001(N=663) . . . . . . . 44 Table 3.3 Industry 
omposition of water emitting fa
ilities with an average of main variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 Table 3.4 Des
riptive statisti
s of air emitting fa
ilities in 2001 (N=3579) . . . . . . . . 45 Table 3.5 Industry 
omposition of air emitting fa
ilities with an average of main variables 46 Table 3.6 Estimation results of ISO 14001 adoption by water emitting fa
ilities . . . . . 47 Table 3.7 Estimation results of water emissions redu
tions . . . . . . . . . . . . . . . . 48 Table 3.8 Estimation results of negative binomial regressions for grievan
es against pol- lution (water emitting fa
ilities) . . . . . . . . . . . . . . . . . . . . . . . . 49 Table 3.9 Estimation results of ISO 14001 adoption by fa
ilities emitting into air . . . . 51 Table 3.10 Estimation results of air emissions redu
tions (SAR) . . . . . . . . . . . . . 52 Table 3.11 Estimation results of air emissions redu
tions (SAR and SAC) . . . . . . . . 53 Table 3.12 Estimation results of negative binomial regressions for grievan
es against pol- lution (air emitting fa
ilities) . . . . . . . . . . . . . . . . . . . . . . . . . . 54 Table B.1 Estimation results of ISO 14001 adoption by water emitting fa
ilities with different 'm's . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 Table B.2 Estimation results of water emissions redu
tions with different 'm's . . . . . . 104 vii Table B.2 Estimation results of ISO 14001 adoption by air emitting fa
ilities with differ- ent 'm's . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 Table B.4 Estimation results of air emissions redu
tions with different 'm's . . . . . . . 106 viii List of Figures Figure 2.1 De
ision tree of game . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Figure 4.1 Cost savings under the tax and the quota when the slope of MAC is steeper (left) and atter (right) than the slope of MD . . . . . . . . . . . . . . . . . 66 Figure 4.2 A 
ase in whi
h a tax separating equilibrium exists . . . . . . . . . . . . . . 70 Figure 4.3 A 
ase in whi
h no tax separating equilibrium exists . . . . . . . . . . . . . 70 Figure 4.4 So
ial 
osts, industries' 
osts, and politi
al 
ontributions when 
= 1=3 . . . 79 Figure 4.5 So
ial 
osts, industries' 
osts, and politi
al 
ontributions when 
= 1 . . . . 80 Figure 4.6 So
ial 
osts, industries' 
osts, and politi
al 
ontributions when 
= 3 . . . . 81 Figure 4.7 Separating equilibrium quotas for the high-
ost industry with the low-
ost indsutry with different levels of natural emissions(ē H = 60,70,and 80) . . . . 84 Figure 4.8 Separating equilibrium tax rate for the low-
ost industry with high-
ost ind- sutry with different levels of natural emissions(ē H = 60,70,and 80) . . . . . 84 ix A
knowledgements I would like to express my gratitude to my supervisory 
ommittee, Brian Copeland, Sumeet Gulati and Werner Antweiler for their support at ea
h stage of this thesis. I am grateful to Brian Copeland and Sumeet Gulati for their suggestions and en
ouragements, whi
h led me to think more deeply about the subje
t and helped me 
omplete the theoreti
al 
hapters of this thesis. In addition, I warmly thank Werner Antweiler for his valuable advi
e. His extensive dis
ussions 
on
erning my work have been very helpful for an empiri
al 
hapter of this thesis. I would also like to thank my external examiner Per Fredriksson for 
arefully reading this dissertation and making useful 
omments. The support of fellow students and friends over the years is deeply a
knowledged with appre
iation. Finally, I also would like to thank my wife Noriko and my parents Yasuhide and Tsurue for their support over the years. x Chapter 1 Obje
tive and Overview 1.1 Obje
tive In the 1960s and 1970s, the beginnings of the environmental movements in developed 
ountries were asso
iated with the establishment of environmental administrations and with the introdu
tion and amendment of national environmental legislation. The approa
hes to pollution 
ontrol dur- ing this period were mainly 
ommand-and-
ontrol and emission standard regulations. From the viewpoint of polluting rms, their environmental responsibility during this period meant regulatory 
omplian
e. However, e
onomists have urged the use of market-based instruments for pollution 
ontrol instead of 
ommand-and-
ontrol regulations be
ause 
ommand-and-
ontrol approa
hes are less ef
ient than market-based instruments and do not provide dynami
 in
entives for te
hnologi
al innova- tion and its diffusion. In addition, politi
ians and environmentalists re
ognized the need for 
ost- effe
tive instruments be
ause pollution 
ontrol 
osts were in
reasing. Therefore, market-based instruments were introdu
ed in the 1980s and 1990s. Tradable permit systems were introdu
ed in the US, while environmental taxes were introdu
e in European 
ountries. More re
ently, voluntary approa
hes, 
ommitments to improve environmental performan
e beyond legal requirements, have played a prominent role in addressing environmental 
hallenges. One rea- son for the popularity of voluntary approa
hes from the supply side of environmental regulations is that it is politi
ally dif
ult to pass environmental laws, espe
ially in the US. However, affe
ted rms started to ta
kle environmental issues not only be
ause of regulation but also in light of poten- tial 
ost redu
tion, liability 
on
erns, and the indire
t impa
t of regulations in the 1980s. Therefore, 1 governments have offered voluntary environmental programs (VEPs) and en
ouraged rms to take voluntary environmental a
tions. For example, the US Environmental Prote
tion Agen
y (EPA) has initiated over one hundred VEPs sin
e the 33/50 Program was laun
hed. Another example involves Japanese lo
al governments that have provided support programs for the adoption of en- vironmental management system (EMS) 
erti
ates, su
h as ISO 14001. These support programs may have 
ontributed to the fa
t that the number of 
erti
ations in Japan was the highest in the world until 2008. This thesis will examine environmental poli
ies, voluntary approa
hes in parti
ular in 
hapters 2 and 3, and emission quotas and market-based instruments in 
hapter 4. Voluntary programs for emissions redu
tions will be studied in 
hapter 2. Voluntary a
tions by Japanese fa
ilities will be examined in 
hapter 3. In 
hapter 4, we 
ompare emission taxes and quotas. Chapter 2 uses a theoreti
al model to explain why governments implement voluntary emission redu
tion programs (VPs). The 
hapter also examines the properties of the VP that the regulator sets to maximize so
ial welfare. In the model, there are 3 types of players: multiple polluting rms, a regulator, and a legislator. The regulator has two options to generate emissions redu
tions from rms: a mandatory standard and a VP. The mandatory standard 
an for
e all rms to redu
e their emissions. However, the approval of the legislator who is affe
ted by the lobby group of the polluting rms is ne
essary to implement it. However, the regulator 
annot for
e rms to parti
ipate in the VP, but legislator approval is not ne
essary to implement it. We show that the VP 
an generate less so
ial 
ost and more aggregate abatement than the mandatory standard. Therefore, the regulator implements the VP when it 
an generate higher so
ial welfare than the mandatory standard. We also nd that 
hanges in parameters affe
t aggregate abatement under the VP more than under the mandatory standard be
ause su
h 
hanges affe
t the abatement rate of individual rms and the parti
ipation rate. Chapter 3 empiri
ally investigates reasons why fa
ilities voluntarily a
quire an environmental man- agement system (EMS) 
erti
ation and the impa
ts of EMS 
erti
ation on environmental perfor- man
e in Japan. We fo
us on ISO 14001 adoption and its impa
ts on environmental performan
e by Japanese fa
ilities during 2001–2003. A

ording to a survey of the Organisation for E
onomi Co-operation and Development (OECD), many fa
ilities see the fa
t that similar fa
ilities adopt en- 2 vironmental pra
ti
es as an important motivation for adopting them. Thus, intra-industry spillover of the adoption of environmental pra
ti
es is likely to exist. Using estimation methods of spa- tial e
onometri
s, we estimate ISO 14001 adoption and its impa
ts on environmental performan
e while 
ontrolling for su
h spillovers and examining the magnitude of spillovers. We show ndings indi
ating that fa
ilities emitting pollution into water are more likely to adopt ISO 14001 iffa
ili- ties that belong to rms with similar revenue in the same industry adopt it and that the per
entage 
hange in emissions into air is 
orrelated between similar-sized fa
ilities in the same industry. Chapter 4 
ompares taxes and quotas when a polluting industry with private information on emis- sion abatement 
osts politi
ally inuen
es the taxes and quotas through lobbying a
tivity. We employ an informed prin
ipal model developed by Maskin and Tirole (1992). We examine how taxes and quotas affe
t a polluting industry's in
entive to inuen
e regulation and how private in- formation affe
ts so
ial welfare and the politi
al inuen
e of the polluting industry. We show that private information 
an improve the so
ial welfare under taxes but 
annot improve it under quo- tas. Private information also redu
es a 
omparative disadvantage of taxes over quotas when the government does not 
are about so
ial welfare. 1.2 Overview This thesis is organized as follows. Chapter 2 presents a theoreti
al analysis of voluntary emission redu
tion programs. Chapter 3 examines the impa
ts of ISO 14001 on environmental performan
e and the intra-industry spillovers of ISO 14001 adoption and environmental performan
e in Japan. Chapter 4 
ompares taxes and quotas when a polluting industry with private information on abate- ment 
osts engages in lobbying. Chapter 5 
on
ludes this thesis. Proofs of the propositions in Chapter 2 are given in Appendix A. Appendix A also in
ludes arguments on the equilibrium abate- ment rate and the number of parti
ipating rms when the equilibrium number of parti
ipating rms derived in proposition 2 is not an integer. In appendix B, we show how to 
al
ulate some tax rate and 
ontribution used in 
hapter 4, and we prove propositions of 
hapter 4. 3 Chapter 2 Why Regulators Adopt Voluntary Programs: A Theoreti
al Analysis of Voluntary Pollutant Redu
tion Programs 2.1 Introdu
tion Voluntary approa
hes to environmental prote
tion have be
ome prominent sin
e the late 1980s. In the European Union (EU), the number of new voluntary agreements in
reased from 6 in 1981 to more than 45 in 1995 (OECD (1999)). There are more than 300 negotiated agreements between governments and polluting industries or rms in Europe. In the United States (US), Brouhle et al. (2005) identied over 50 voluntary programs at the federal level alone sin
e 1991, when the 33/50 program 1 , the rst voluntary program, was laun
hed. In Japan, over 30,000 negotiated agreements between lo
al governments and polluters are in effe
t 2 . Voluntary approa
hes 
an take various forms. However, based on government involvement, we 
an 
ategorize them into three types. The rst type is a unilateral a
tion that is initiated and 
arried out by rms and industries. The se
ond form is a bilateral agreement between the government and 1 The US EPA laun
hed this program to redu
e aggregate emissions of 17 
hemi
als by 33% in 1992 and by 50% in 1995, relative to the 1988 baseline. We will exaplain more in se
tion 2. Or see Khanna Khanna (2007) for a more detailed review of the 33/50 program. 2 The reason why over 30,000 agreements are being implemented in Japan is that most agreements have been between a rm and muni
ipality. When a rm 
onstru
ts (or extends) its fa
ility, it 
on
ludes an agreement with a muni
ipality at whi
h the fa
ility is (will be) lo
ated. This is a typi
al setting in whi
h agreements are made. Please see Wel
h and Hibiki (2003) for details. In 
ontrast to Japan, many agreements o

ur at the federal government or industrial level in Europe and the US. This is why the number of agreements or voluntary poli
ies is so different between Japan and other developed 
ountries. 4 industries or rms. Both the government and polluters a
tively set the target and other parameters for the approa
h. The nal type and the fo
us of this 
hapter is a voluntary program (VP) designed by a government. The government sets the obje
tives of the program, and the rms have a right to de
ide whether they will parti
ipate in the program. Thus, parti
ipation in these programs is “voluntary”. Be
ause rms 
an 
hoose not to join a VP, it is subje
t to the free-rider problem, whi
h means that the VP is implemented but not undertaken by all rms. Therefore, parti
ipation rate is an important 
riterion when we evaluate VPs. Parti
ipation rates for some VPs are very low. Parti
ipation rates for the US 33/50 program and the Canadian ARET program, whose goal is to redu
e the release and/or transfer of 
hemi
als, are 17.0% and 13.4%, respe
tively. Relative to its parti
ipation rate, the total release and transfer of parti
ipating rms in 1988 (baseline year) was high (62.5%) 3 . However, the program did not 
over about 40% of total releases and transfers. Why do VPs 
ontinue to be implemented even when parti
ipation rates are low? Politi
al dif- 
ulty in passing environmental laws is likely to 
ontribute to the in
reased popularity of voluntary approa
hes, and therefore, it must be one of the reasons for the 
ontinued implementation of VPs despite low parti
ipation rates. As an example, a 
arbon tax was proposed at the EU level but failed due to industrial lobbying. Just after this tax failed to pass in the EU, many EU 
ountries adopted voluntary approa
hes to 
limate 
hange. Thus, governments adopt voluntary approa
hes instead of mandatory regulations in response to politi
al pressure from polluting industries. Moreover, when a pollution problem is not a hot politi
al issue, it is politi
ally dif
ult to implement effe
tive poli
ies. Parti
ipation rate of VPs would likely be low in su
h 
ases. In addition to these politi
al dif
ulties, governments might sa
ri
e high parti
ipation rates for the sake of effe
tiveness. A government might have to set low abatement rates for ea
h parti
ipating rm to maximize parti
ipation, and, as a result, aggregate abatement may be low. Thus, a high parti
ipation rate does not always ree
t the effe
tiveness of the VP, and we 
annot evaluate a VP based on the parti
ipation rate alone. However, the abatement rate of ea
h parti
ipating rm is not an appropriate solitary 
riterion for evaluating a VP either. We have to take into a

ount both rates 3 Many of ARET substan
es were not required to be reported to the National Pollutant Release Inventory (NPRI), whi
h legally mandates publi
 reporting. Therefore, data on these substan
es are not available. See Antweiler and Harrison (2007) for NPRI-re
orded emissions of ARET-listed substan
es and ARET-parti
ipating rms' share of them. 5 for ea
h parti
ipating rm when evaluating a VP. In addition, examining the relationship between parti
ipation and abatement rates in effe
tive VPs provides useful information for the design of new VPs. This 
hapter aims to explain why governments implement VPs even though their parti
ipation rates 
an be low and to examine the relationships between parti
ipation rates, requirements (abatement rates) and the ef
ien
y of VPs. To do so, we built a model with three types of players: a regulator, legislator, and multiple polluting rms. In the model, the regulator is assumed to be benevolent be
ause it attempts to maximize so
ial welfare. Two regulation options, a mandatory standard and a VP, are available to the regulator to redu
e emissions. For the regulator to implement the mandatory regulation, the legislator, who is inuen
ed by a lobby group of polluting rms through politi
al 
ontributions, must approve it. This legislative pro
ess of setting mandatory standards is intended to 
apture the politi
al dif
ulty involved in setting the mandatory poli
y, as dis
ussed above. In general, the mandatory standard is inef
ient. Hen
e, the regulator might have an in
entive to offer the VP to a
hieve more ef
ient abatement allo
ation, although the regulator 
annot for
e the rms to parti
ipate. In addition to the regulator, the individual rms might have an in
entive to parti
ipate in the VP to save lobbying 
osts. Along with a sharp in
rease in the number of voluntary approa
hes, there has also been an in
rease in resear
h on these approa
hes. Several papers have theoreti
ally investigated voluntary agree- ments (VAs), and Lutz et al. (2000) and Maxwell et al. (2000) studied unilateral a
tions 4 . Lyon and Maxwell (2003) examined the effe
ts of poli
y 
hoi
e, taxes or VAs on the adoption of new te
hnology when rms unilaterally adopt 
lean te
hnology. Be
ause governments design VPs but do not design the unilateral a
tions by polluters, the models and the results of these studies on unilateral a
tions 
annot be applied to VPs. Most papers on VAs have built and analyzed single polluter models (e.g., Segerson and Mi
eli (1998), Hansen (1999), Segerson and Mi
eli (1999), 4 Following the publi
ation of the seminal papers of Segerson and Mi
eli (1998), Hansen (1999), and Segerson and Mi
eli (1999) who analyzed VAs, Manzini and Mariotti (2003) studied VAs between a regulator and group of heterogeneous rms. Gla
hant (2007) examined Vas, whi
h 
annot enfor
e a rm's (or an industry's) 
ommitments by a model with a polluting industry, regulator, and legislator. Lutz et al. (2000) studied the effe
t of a “green” 
ommitment by rms when their a
tions inuen
e future regulation. Maxwell et al. (2000) showed the possibility of self-regulation preempting future regulation. The model of Lutz et al. (2000) and Maxwell et al. (2000) and its impli
ations were explained Lyon andMaxwell (2004). See also Lyon andMaxwell (2002) for a survey of early theoreti
al and empiri
al studies of voluntary approa
hes. 6 and Gla
hant (2007)). Even with multiple polluters, all rms are assumed to 
olle
tively negotiate the VA with the government, as in Manzini and Mariotti (2003), or the voluntary agreement is not a subsidy for the adoption of te
hnology (Lyon and Maxwell (2003) ). Thus, most theoreti
al pa- pers have not in
orporated an individual rm's parti
ipation de
ision into voluntary programs or agreements with emission redu
tion targets, and, therefore, they have not explained why VPs with low parti
ipation rates are implemented. The work of Dawson and Segerson (2008) is an ex
eption in the literature. Dawson and Segerson (2008) developed models in
orporating the parti
ipation of individual rms into the voluntary pro- gram or agreement. They examined the 
ase in whi
h the regulator has two options to a
hieve an (exogenous) aggregate emission level target via a VP or an emission tax. They analyzed the existen
e and properties of an equilibrium parti
ipation rate and its so
ial welfare impli
ations. However, there are no reasons that the regulator would offer a VP be
ause it is always “not bet- ter” than the tax, and the regulator 
an always 
hoose the tax. In 
ontrast, the VP is implemented be
ause the regulator preferred it in this 
hapter be
ause of the presen
e of two publi
 agents, a benevolent regulator and rent-seeking legislator, similar to the model des
ribed by Gla
hant (2007). If the benevolent regulator sets a mandatory poli
y, su
h as an emission tax, the poli
y is 
ost- effe
tive. Therefore, the regulator has no in
entive to implement the VP. However, the VP might generate higher so
ial welfare than the mandatory poli
y if the rent-seeking legislator is the entity that would set the mandatory poli
y. We show that the regulator 
an implement the VP, whi
h generates less so
ial 
ost and more ag- gregate abatement than the mandatory standard. By adopting the VP, the regulator 
an make par- ti
ipating rms allo
ate resour
es that would otherwise be used to lobby for emissions abatement. The regulator's problem is maximizing the resour
es that are reallo
ated from lobbying efforts to emissions abatement, subje
t to the 
onstraint that there is no new parti
ipation if a portion of the abatement from the parti
ipating rms is greater than the aggregate abatement under the mandatory poli
y. In this 
ase, the resour
e reallo
ated to abatement is maximized by making the “newest” parti
ipating rm abate as mu
h as possible and putting the parti
ipation rate aside. Thus, in the most ef
ient VP, there are some “non-parti
ipating” rms. Moreover, we show that a VP with a low parti
ipation rate 
an generate less so
ial 
osts and more aggregate abatement than the manda- 7 tory standard if it is politi
ally dif
ult to set the mandatory standard. The other results of this study are as follows. Changes in parameters affe
t aggregate abatement under the VP through two 
hannels, the parti
ipating rate and the individual parti
ipating rms' abatement rates, but aggregate abatement under the mandatory poli
y is affe
ted by the abatement rate of the individual rms alone. Therefore, 
hanges in parameters affe
t aggregate abatement levels under the VP more than they do under the mandatory standard. If the mandatory poli
y is stringent, the VP is effe
tive and its parti
ipation rate is high. For instan
e, when there are many polluters or ea
h polluter has high emissions, the mandatory poli
y is stringent be
ause aggregate emissions are large in su
h 
ases. Therefore, rms in a more pollution-intensive industry are more likely to join su
h a VP. The parti
ipation rate is also high when there are many polluters (if other parameters are the same). However, rms in an industry with higher abatement 
osts are less likely to join be
ause the mandatory poli
y is less stringent. Although the regulator implements a low parti
ipation rate VP when it is politi
ally dif
ult to implement a mandatory poli
y, it is also possible that most or all rms will parti
ipate in the VP. Many rms parti
ipate in the VP when its net benet is greater than that of the mandatory standard. These results are totally different from that of international environmental agreement (IEA) in Barrett (1994) who argued that polluters are free to join IEAs or not, mu
h like a VP. The mandatory standard, a kind of punishment for non-parti
ipation to all polluters, 
reates the differen
e between VPs and IEAs. This 
hapter is organized as follows. In next se
tion, we review fa
ts on parti
ipation in voluntary approa
hes. Se
tion 3 presents a model and analyzes situations of equilibrium. We present a 
omparative stati
s analysis in Se
tion 4. Se
tion 5 dis
usses the results of 
hanging the settings from a basi
 setting. Se
tion 6 
on
ludes this 
hapter. 2.2 Some fa
ts on voluntary programs In this se
tion, we review some fa
ts on parti
ipation in voluntary approa
hes based on the 33/50 program, the ARET program and environmental agreements in Europe. We then review the politi
al 8 Table 2.1: Overview of 33/50 program (Sour
e: US Environmental Prote
tion Agen
y (1999)) 1st round 2nd round 3rd round Not 
on- ta
ted Conta
ted total Number of 
ompanies Companies with 33/50 fa
ilities 504 4534 2512 2612 7550 10167 Parti
ipating 
ompanies (PCs) 328 819 140 7 1287 1294 Parti
ipation rate 65.1% 18.2% 5.6% 0.3% 17.2% 12.7% Quantities of 17 
hemi
als (millions of pounds) Total releases & transfers in 1988 993 367 45 89 1405 1496 Total from PCs 809 110 14 1 933 935 Share of releases & transfers from PCs 81.5% 30.0% 31.1% 1.1% 66.4% 62.5% pro
ess of adopting voluntary approa
hes. The US 33/50 program was laun
hed by the Environ- mental Prote
tion Agen
y (EPA) in 1991. The goal of this program was to redu
e the release and transfer of 17 
hemi
als by 33% in 1992 and by 50% in 1995, relative to the 1988 baseline. The EPA sent letters to the CEOs of the parent 
ompanies of the emitting fa
ilities to en
ourage parti
- ipation. In January 1988, the “top” 600 
ompanies, whi
h a

ounted for 66% of the total release of these 
hemi
als in 1988, were invited to parti
ipate, and 5,400 other 
ompanies were invited in July 1991. After 1992, the EPA invited 2,512 
ompanies that began emitting the aforementioned 17 
hemi
als after 1988 to parti
ipate. Table 2.1 gives an overview of the 33/50 program. The parti
ipation rate of the invited 
ompanies is 17.0%, and the share per
entage of the total release and transfer of all substan
es by the rms invited in 1988 was 62.5%. Clearly, the parti
ipation rate for the rst round of invited 
ompanies was mu
h higher (65.1%) than it was for the 
ompanies invited later. There have been several empiri
al studies on the 33/50 program. For example, Gamper-Rabindran (2006) found that plants in the 
hemi
al industry, whi
h uses hazardous air pollutants intensively, were more likely to parti
ipate. She also found that the parti
ipation rate of rms in the 
hemi
al industry in
reased when the past inspe
tion rate was high. Innes and Sam (2008) found that large rms listed as potentially-responsible parties for a large number of Superfund sites were more likely to join. 9 The Canadian A

elerated Redu
tion/Elimination of Toxi
s (ARET) program was laun
hed in 1994. ARET targeted 117 toxi
 substan
es, in
luding 30 substan
es that persist in the environ- ment and may a

umulate in living organisms. ARET required all rms to voluntarily redu
e their release of those 30 substan
es by 90% and 87 other substan
es by 50% by the year 2000. This program was guided by a 
ommittee of stakeholders established in 1992, whi
h in
luded represen- tatives from industry, health and professional asso
iations, in addition to representatives from the federal and provin
ial governments. Although the ARET program has some 
hara
teristi
s of a ne- gotiated voluntary agreement, it has more in 
ommon with the 33/50 program than with negotiated agreements be
ause it 
alled on all rms to redu
e their emissions. Antweiler and Harrison (2007) found that (1) large and emission-intensive fa
ilities were more likely to parti
ipate; that (2) parti
ipation rates in
reased along with the parti
ipation of other fa
ilities in the same industry (spillover within industry); and that (3) the parti
ipation rate for members of the trade asso
iations that were involved in negotiating the terms of ARET was higher than for other fa
ilities. The parti
ipation of rms 
an be an important issue for negotiated voluntary agreements and for publi
 voluntary programs. The European Environment Agen
y (1997) surveyed voluntary agree- ments in Europe and found that many small and medium enterprises (SMEs) tended to be free-riders when trade asso
iations were dire
tly involved with the agreements. In addition, the European En- vironment Agen
y (1997) admitted that free-riding by SMEs might jeopardize agreements even though it may not affe
t their ef
ien
y. Hen
e, voluntary approa
hes to 
ontrolling pollution from multiple rms are subje
t to the problem of free-riders, whi
h is one of the important issues to be addressed. Politi
al fa
tors play a role in the adoption of voluntary approa
hes to environmental prote
tion. Many voluntary programs and agreements have been adopted without approval from a legislative se
tor. For example, the EPA founded the 33/50 program, and the program was implemented without the approval of the legislative se
tor. In addition, in response to politi
al pressure from polluting industries, some voluntary approa
hes have been adopted as an alternative to mandatory poli
ies. For example, many EU 
ountries adopted VAs in the mid-1990s after the EU 
arbon tax proje
t was abandoned due to pressure from lobby groups for the energy-intensive industries. 10 Another example is the Clinton administration's Climate Change A
tion Plan (CCAP). Although the administration originally proposed an energy tax, it adopted voluntary programs to meet the goals of the CCAP due to politi
al opposition from industries. Our model in
ludes two types of publi
 agen
ies, a legislative se
tor and a regulator, to 
apture this type of politi
al pro
ess. 2.3 The model The following three types of players are involved in our model: a regulator, a legislator and N polluting rms. The regulator is benevolent in that he/she aims to minimize so
ial 
osts, the sum damage by pollution, and aggregate abatement 
osts. In the rst stage, the regulator announ
es a voluntary a(%) emission redu
tion program (VP). In the se
ond stage, polluting rms de
ide whether to parti
ipate in the VP or not. Thus, the rms do not have to parti
ipate in the VP, but the regulator asks the legislator to ena
t a mandatory emission redu
tion standard if an insuf
ient number of rms parti
ipate. When the mandatory redu
tion standard is set or the insuf
ient number of rms parti
ipate in the VP, polluting rms will 
olle
tively lobby against it and then the legislator will set the mandatory standard. Therefore, the mandatory standard will typi
ally be so
ially inef
ient. On the other hand, if an suf
ient number of rms parti
ipate in the VP in the se
ond stage, then the VP is implemented. The timing of this model is des
ribed in Figure 2.1. It is possible that the polluting rms will take voluntary a
tion before the regulator announ
es the VP. However, if many rms are involved, it is hard for trade asso
iations or rms to 
oordinate voluntary emission redu
tion efforts. This 
hapter 
onsiders su
h a 
ase. Usually, polluting rms unite in opposition to mandatory regulations, although not all rms join su
h 
olle
tive a
tions. In su
h a situation, the polluting rms form a lobby group, whi
h will inuen
e mandatory regulations through politi
al 
ontributions nan
ed by voluntary 
ontributions from the rms. This 
hapter does not address the possibility that a legislator sets a mandatory poli
y even if a VP is already in pla
e. However, to 
reate a law that enfor
es the mandatory poli
y, various kinds of information, su
h as s
ienti
 knowledge on pollution, must be obtained and integrated. In many 
ases, it is likely logisti
ally dif
ult and/or 
ostly to obtain su
h te
hni
al information and exper- tise without the 
ooperation of a government-run environmental organization, whi
h is represented 11 Figure 2.1: De
ision tree of game by the regulator in this 
hapter. Therefore, it is not unreasonable to assume that the regulator would initiate both poli
ies. Figure 2.1 shows that the regulator has two options: implement the mandatory poli
y or the vol- untary program to minimize so
ial 
osts, whi
h are the sum of damage by pollution and aggregate abatement 
osts. We assume that the damage is a quadri
 fun
tion of emission, 1 2 d( å i (1a)ē i ) 2 , and that the abatement 
osts of a rm i is linear in emission, 
a ē i where ē i , a , 
 and d are the pollutant emission of rm i before regulation, the redu
tion rate, marginal abatement 
ost and the slope of marginal damage, respe
tively. Be
ause the regulator is benevolent, the regulator's 
ost under the mandatory poli
y is the sum of the aggregate abatement 
osts and damages R M (a) = 1 2 d ( N å i ((1a)ē i ) ) 2 + N å i 
a ē i : (Madantory poli
y) If the regulator 
ould set a mandatory standard or a redu
tion rate, he/she would set the redu
tion rate a  = 1  dē i N : 5 (2.1) 5 We have this from the F.O.C for minimization of R M (a). We fo
us on a 
ase that (2.1) is nonnegative. Otherwise, we do not need to implement regulation. 12 However, the legislator sets the redu
tion rate for the mandatory program. When the 
ongress en- a
ts the mandatory standard, the lobby group of polluting rms inuen
es it by offering politi
al 
ontributions to the legislator, who is a representative or median legislator. As Gla
hant (2007) de- s
ribed, the legislator's payoff fun
tion is assumed to be a weighted sum of politi
al 
ontributions, W and so
ial welfare (negative so
ial 
osts), L(a;W) = lW (1l )[ 1 2 d( N å i (1a)ē i ) 2 + N å i 
a ē i ℄ where l 2 (0;1). l 
an be interpreted as the responsiveness of the legislator to lobbying or the politi
al dif
ulty of setting an ef
ient mandatory abatement rate. Politi
al dif
ulty may o

ur when a pollution problem is not an important issue on the politi
al agenda. A high l may ree
t that it is not an important issue on the agenda. Politi
al 
ontributions from the lobby group are nan
ed by voluntary 
ontributions from individ- ual rms. Ea
h rm is assumed to make a 
ontribution to the lobby group taking other rms' 
ontributions as given like voluntary publi
 good provision games. Firm i's 
ost is the sum of abatement 
osts, 
a ē i , and 
ontributions to the lobby group, W i , 
a ē i +W i . In 
ontrast to the mandatory poli
y, the regulator 
an set the abatement rate of the VP, but it 
annot for
e rms to parti
ipate. Hen
e, if the regulator implements the VP, the obje
tive is min a R V (a) = 1 2 d ( N P å i ((1a)ē i )+ NN P å j ē j ) 2 + N P å i 
a ē i (VP) where N P is the number of rms parti
ipating in the VP. As mentioned above, individual rms 
an de
ide whether they parti
ipate in the VP. Firms have no in
entive to join the VP if their 
ost under the VP is greater than it would be under the mandatory poli
y. Therefore, N P and the VP parti
ipation rate depend on the redu
tion rate of the VP and on ea
h rm's 
osts or redu
tion rate under the mandatory standard. In the next subse
tion, we analyze how the mandatory poli
y is set. Then, we examine how the regulator sets the VP. 13 2.3.1 The legislative subgame First, we analyze the 
ase in whi
h the regulator uses the government to set the mandatory standard. In the nal stage, the legislator sets the mandatory redu
tion rate, and rms redu
e their emissions. Before the nal stage, the lobby group offers politi
al 
ontributions depending on the redu
tion rate a . Be
ause the legislator 
an reje
t the offer from the lobby group and 
hoose a so
ially optimal redu
tion rate, the potential politi
al 
ontribution must satisfy L(a;W) L(a  ;0). This 
onstraint must hold equally be
ause the total 
ost to the rms in
reases with in
reases in the redu
tion rate, and the lobby moves rst to make a non-negotiable offer. Therefore, politi
al 
ontributions and the redu
tion rate have the following relationship W(a) = 1l l " 1 2 d( å i (1a)ē i ) 2 + å i 
a ē i  [ 1 2 d( å i (1a  )ē i ) 2 + å i 
a  ē i ℄ # : (2.2) Let ˆ a(W) be the redu
tion rate satisfying (2.2) when politi
al 
ontribution is W. Using ˆ a(W), we 
an des
ribe an individual rm's problem. Their 
ontributions affe
t politi
al 
ontributions dire
tly and redu
tion rates indire
tly. Therefore, taking it as a given that other rms have also made 
ontributions, rm i 
hooses its 
ontribution to minimize its total 
ost W i + ˆ a(W i + å j 6=i W j )ē i ; (2.3) subje
t to (2.2). Using W= W i + å j 6=i W j , we enter (2.2) into (2.3) and take the derivative of (2.3) to yield 
ē i ¶ ˆ a ¶W i = 1l l [d å i ē i ( å i (1a)ē i ) å i 
ē i ℄ ¶ ˆ a ¶W i : (2.4) Therefore, the abatement rate 
hosen via the legislative pro
ess is a L = 1  dNē  
l dN 2 ē(1l ) : 6 (2.5) 6 In the following analysis, we assume a L is positive. Abatement rate a L should be 0 if a L is negative. A 
ontribu- 14 The third term of RHS is the differen
e in the redu
tion rates between the so
ially optimal and the mandatory standard. The absolute value of this differen
e de
reases with the number of polluting rms be
ause politi
al 
ontribution de
reases due to free-riding. Therefore, if the so
ially optimal redu
tion rate is the same, the more rms pollute and the more effe
tive the mandatory standard is. If we fo
us on a symmetri
 equilibrium, then rm i's 
ontribution is W i =W(a L )=N = 1l l " 1 2 d( å i (1a L )ē i ) 2 + å i 
a L ē i  [ 1 2 d( å i (1a  )ē i ) 2 + å i 
a  ē i ℄ # =N =  2 l 2dN 3 (1l ) (2.6) Finally, an individual rm's 
ost under the mandated standard is ¯ C = 
a L ē+W i = 
ē  2 dN   2 l dN 2 (1l ) +  2 l 2dN 3 (1l ) : (2.7) 2.3.2 The VP subgame If polluting rms parti
ipate in the VP, then their 
osts must be smaller than they would be under the mandatory standard. Hen
e, 
a ē i  ¯ C: (2.8) Any parti
ipating rm will not 
hange its parti
ipation de
ision in a state of equilibrium. Ea
h rm makes the de
ision to parti
ipate or not, and therefore rms do not unilaterally 
hange their parti
ipation in a state of equilibrium. Dawson and Segerson (2008) showed that this 
ondition is equivalent to the 
ondition that no parti
ipating rm has an in
entive to unilaterally be
ome a tion is given by W 0 = 1 2 dN 2 ē 2 i N
ē i +  2 2d : If ¯ C 0 =W 0 , then analysis of VP game is the same as in the 
ase where a L is non-negative. 15 non-parti
ipating rm and, te
hni
ally d 2 [(N (N P 1)a)ē i ℄ 2 +(N P 1)
a ē i  R M (a L ): (2.9) This 
ondition means that the regulator will not implement the VP if only N P  1 polluting rms parti
ipate. If this 
ondition does not hold, then the regulator will still implement the VP even if one of the parti
ipating rms does not parti
ipate (i.e., deviate). Thus, parti
ipating rms have an in
entive to unilaterally deviate if this 
ondition does not hold. Subje
t to (2.8) and (2.9), the regulator tries to maximize its payoff or minimize so
ial 
osts. Let a V be su
h that 
a V ē i = ¯ C. When the redu
tion rate under VP is a V , the 
ost of the VP for the parti
ipating rms is the same as the 
osts of the mandatory poli
y. Then, the following lemma holds. Lemma 2.1. The regulator's problem is the same as maximizing aggregate abatement subje
t to (2.8) and (2.9). In addition, we 
an rewrite (2.9) as (N P 1)a ē i  Na L ē i : (2.10) Proof: Due to the linearity of abatement 
osts, the allo
ation of abatement does not affe
t so
ial welfare. Based on (2.7) and (2.8), a V = 1  dNē  
l dN 2 ē(1l ) + 
l 2dN 3 ē(1l ) = 1  dNē  
l 2dN 3 ē(1l ) (2N1) 1  dNē i =a  : Hen
e, under the VP, ea
h parti
ipating rm abates less than the so
ial optimal level, and the greater aggregate abatement results in a smaller so
ial 
ost. Therefore, the regulator's problem is the maximization of the aggregate abatement, whi
h is subje
t to (2.8) and (2.9), and (2.9) is equivalent to (2.10) . QED The regulator's problem is the maximization of the aggregate abatement level, whi
h is subje
t to (2.8) and (2.10). (2.10) implies that (N P  1) parti
ipating rms' abatement levels must be 16 below Na L ē i and independent of N P . Hen
e, maximizing aggregate abatement is equivalent to maximizing the “N P th” parti
ipating rm's abatement, and the following proposition holds. Proposition 2.1. The largest aggregate abatement under the VP is generated by the abatement rate a V and the parti
ipation rate, N V P =N, N V P =N = 8 > < > : a L =a V +1=N If N > a V =(a V a L ) 1 Otherwise. Proof: If (N 1)a V ē i  Na L ē i or N  a V =(a V a L ), then all rms have an in
entive to join the VP be
ause (2.8) and (2.10) hold for any N P  N. Be
ause a V is the highest redu
tion rate, a V with N V P =N = 1 gives the highest aggregate abatement. If N > a V =(a V a L ), (2.10) must hold with equality to maximize the aggregate abatement. So, By substituting (N V P  1)a ē= Na L ē into N V P a ē, we have N V P a ē= a ē+(N V P 1)a ē= a ē+Na L ē. Be
ause Na L ē is set by the 
ongress, the maximization of N V P a ē is equal to that of a ē or a . Therefore, the redu
tion rate whi
h generates the largest aggregate abatement is the highest a that satises (2.8) and whi
h is a V . In addition, we have N V P = Na L =a V +1 from (2.10). QED See Appendix A.1 for the 
ase where the N V P is not an integer. Two fa
tors 
an be manipulated to in
rease the aggregate emission abatement or de
rease so
ial 
osts: the parti
ipation rate and the redu
tion rate. The regulator prefers high abatement and parti
ipation rates, but a trade-off exists between the redu
tion rate and the parti
ipation rate due to Constraint (2.10) whi
h is a kind of a 
onstraint on aggregate abatement. (2.10) means that the abatement by N P  1 rms must not be greater than the aggregate abatement level under the mandatory poli
y regardless of the parti
ipation rate. By substituting (N P 1)a ē i = Na L ē i ((2.10) with equality) into N V P a ē, we have N V P a ē = a ē+Na L ē. Be
ause Na L ē is set by the 
ongress, the largest aggregate abatement is a
hieved by 
hoosing the highest redu
tion rate su
h that (2.8) holds. This proposition means that the VP is the most environmentally effe
tive if the regulator 
hooses the highest abatement rate, su
h that the parti
ipating rms' abatement 
osts under the VP are less than under the mandatory poli
y. If the regulator does 
hoose the highest abatement rate, then some 17 rms generally do not join the VP. Thus, symmetri
 rms may take asymmetri
 a
tions (i.e., some of them parti
ipate, but others do not). Of 
ourse, all the rms might parti
ipate in some 
ases 7 . In addition, the proposition implies that the regulator might implement the VP even if the parti
ipation rate is low (e.g., when the legislator does not 
are about the so
ial 
osts related to pollution or when it is politi
ally dif
ult to set the mandatory poli
y). In the next se
tion, we dis
uss how 
hanges in parameters (redu
tion rate, parti
ipation rate and aggregate abatement) affe
t the VP when it is effe
tive. To understand why symmetri
 rms might take asymmetri
 a
tions and what happens as a result, we 
onsider simple 
ases with two rms. The payoffs for the rms and the regulator are des
ribed in Table 2.2. The left, 
enter and right values in ea
h 
ell represent the payoffs of Firm 1, Firm 2 and the regulator, respe
tively. P and NP stand for “parti
ipate” and “not parti
ipate”, respe
tively. The payoffs for rms in the 
ase when both rms do not parti
ipate (NP, NP) are equal to the payoffs when the mandatory poli
y is implemented be
ause the regulator prefers the mandatory poli
y in su
h a 
ase. However, the rms' payoffs in the other 
ases are the same as they are when the VP is implemented. The set of payoffs is likely to be similar to those in the left matrix when the legislator is not 
on
erned with so
ial welfare, whi
h is low under the mandatory poli
y is low when the legislator 
ares less about it than about politi
al 
ontributions. However, payoffs are similar to those in the right matrix when the legislator 
ares about so
ial welfare. It should be noted that a parti
ipating rm re
eives the same payoff under the VP as it would under the mandatory poli
y if the redu
tion rate is a V (from (2.8)) . Table 2.2: Payoff matri
es when some rms parti
ipate (left) and when all rms parti
ipate (right). rm2 rm2 P NP P NP rm1 P 0,0,1 0,1,0.5 rm1 P 0,0,0.5 0,1,-0.5 NP 1,0,0.5 0,0,0 NP 1,0,-0.5 0,0,0 Consider (P, NP) in the left matrix. Be
ause the regulator's payoff is greater than it would be in (NP, NP), the regulator will implement the VP. Given the parti
ipation of Firm 2, Firm 1 has no in
entive to 
hange from P to NP. In addition, Firm 2 has no in
entive to 
hange from NP to P. 7 For example, the legislator 
ares about so
ial welfare very mu
h, or damage by pollution is serious. 18 Hen
e, (P, NP) is a VP equilibrium. (NP, P) is also a VP equilibrium be
ause the same argument holds. In the right matrix, neither (P, NP) nor (NP, P) is a VP equilibrium be
ause the regulator prefers (NP, NP) over (NP, P) or (P, NP). Hen
e, the regulator will implement the mandatory poli
y, and the payoffs would be the same as with (NP, NP) if either rm does not parti
ipate. In addition, all players weakly prefer (P, P) over (NP, NP), and so they prefer (P, P). Therefore, neither rm has an in
entive to 
hange from P to NP. (P, P) is an equilibrium. However, when no rms parti
ipate in the VP (NP, NP) and the mandatory poli
y is implemented, there is an equilibrium in both examples be
ause the parti
ipation of one rm is irrelevant if the other rm does not parti
ipate. In the right matrix 
ase, there also exists an equilibrium where either rm joins the VP, but the mandatory poli
y is implemented, as dis
ussed above. As we 
an guess, “parti
ipation” versus “non-parti
ipation” is irrelevant to a rm if less than N P  1 of the other rms parti
ipate. Hen
e, there also exists a situation of mandatory regulation equilibrium. In the two examples des
ribed above, the regulator's payoff under the VP equilibrium is the greatest of all the situations of equilibrium. Be
ause the regulator is benevolent, this means that the VP 
an generate higher so
ial welfare than the mandatory poli
y. Moreover, there is a situation of equilibrium where the mandatory poli
y is implemented. The next proposition formally shows these two out
omes. Proposition 2.2. The redu
tion rate a V with parti
ipation rate N V P =N generates higher aggregate abatement and lower so
ial 
ost than a mandatory standard. However, there also exist equilibria where only less than (N V P 1) rms parti
ipate in the VP and the mandatory poli
y is implemented. Proof: When N V P < N, N V P a V ē i > (N V P  1)a V ē i = Na L ē i . If N V P = N, N V P a V ē i > Na L ē i be
ause a V > a L . Lemma 1 implies that a higher aggregate abatement results in lower so
ial 
osts. There- fore, a V with N V P =N generates higher aggregate abatement and lower so
ial 
ost than a mandatory standard. However, the mandatory poli
y generates a higher aggregate abatement if the abatement rate is a V and there is no parti
ipating rm, for example. The mandatory poli
y is implemented in su
h a 
ase. 19 Consider Firm k's parti
ipation de
ision. We suppose that Suppose N V P > 2 and that other rms do not parti
ipate in the VP. It should be re
alled that (N V P 1)a V ē i = Na L ē i . The mandatory poli
y is implemented whether Firm k parti
ipates or not be
ause the parti
ipating rms number fewer than (N V P 1) under both 
ases (rm k parti
ipate or it does not). Thus, rm k has no in
entive to 
hange its parti
ipation de
ision, from “not parti
ipate” to “parti
ipate”. Therefore, there exists an equilibrium where no rms parti
ipate in the VP and the mandatory poli
y is implemented. Please see appendix A.2 for the proof of the general 
ase. QED When the mandatory poli
y is implemented, the rms' 
ost is the sum of the emission abatement 
osts and politi
al 
ontributions. Given that the rms 
annot inuen
e the VP through 
ontribu- tions, they do not spend resour
es lobbying for the VP. Therefore, the regulator 
an design the VP, whi
h makes individual rms in
rease expenditures and redu
e the amount of emissions relative to the mandatory poli
y (subje
t to (2.8) and (2.10)). It is possible that the VP 
an generate higher so
ial welfare than the mandatory poli
y if enough rms join the VP. In 
ontrast with Dawson and Segerson (2008) model, the regulator in this model might prefer the VP over the mandatory standard if the parti
ipation rate is high enough, whi
h depends on the rms' beliefs about other rms' parti
ipation de
isions. This result o

urs due to the differen
e between the regulator's obje
tive and the mandatory poli
y-making pro
ess in this 
hapter versus that des
ribed by Dawson and Segerson (2008). In the latter, the regulator's obje
tive is to a
hieve some aggregate emission level, and in our model, it is to minimize so
ial 
osts. Be
ause some rms might not join the VP, the mandatory standard is better than the VP for a
hieving some aggregate abatement level. Without politi
al dif
ulties, the regulator 
an implement the mandatory poli
y, whi
h a
hieves a so
ially-ef
ient out
ome. Thus, the differen
e between this 
hapter and Dawson and Segerson's paper is 
ru
ial to the implementation of a VP by the government. Furthermore, in our model, the mandatory poli
y is implemented if the polluting rms think that few rms will parti
ipate in the VP. Firms are likely to believe that enough rms will parti
ipate if they feel obliged to prioritize environmental issues in 
orporate-so
ial responsibility (CSR). In the mid-1980s, environmental a
tivists started to in
rease pressure on rms to manage environmen- tal issues in the US. Consequently, the industry undertook environmental initiatives. The 33/50 20 program, mentioned above, was implemented in the early 1990s, when many rms began to 
on- sider environmental issues as their so
ial responsibility 8 . Therefore, the existen
e of situations of equilibrium related to the mandatory poli
y seems 
onsistent with the fa
t that few voluntary approa
hes were implemented prior to the 1990s. The VP is more likely to be implemented if rms feel that they have to ta
kle environmental issues as part of their CSR. Firms think that parti
ipation in the VP appeals to the regulator, even when the VP is not implemented. Therefore, rms gain a small positive benet (smaller than the abatement 
ost) from their 
ommitment to join the VP. However, the regulator does not re
ognize this benet. As long as the parti
ipation rate is low enough that the regulator does not implement the VP (as long as (2.10) holds), ,rms prefer to join be
ause the regulator sets the redu
tion rate (slightly) lower than a V . Therefore, rms get the benet of joining the VP. Thus, a suf
ient number of rms always join the VP. Finally, it is worth mentioning that there is a differen
e in the number of parti
ipating polluters between voluntary programs and international environmental agreements (IEA). Both are 
olle
tive a
tions intended to redu
e emissions, and polluters 
an 
hoose whether or not to parti
ipate. The number of VP-parti
ipating polluters is quite different from that of IEA-parti
ipating polluters (when the abatement 
ost fun
tion is linear and the damage fun
tion is quadrati
). On one hand, Proposition 4 of Barrett (1994) implies that the number of IEA-parti
ipating polluters is always fewer than two. On the other hand, all or most polluters might parti
ipate in a VP, although the number of parti
ipating polluters depends on parti
ular parameters, espe
ially the inuen
e of politi
al 
ontributions on legislators. This differen
e is due to the existen
e of a punishment for non-parti
ipation for all polluters: mandatory regulation. The strength of this punishment depends on how mu
h the legislator is inuen
ed by politi
al 
ontributions. 8 A

ording to survey data, about 50 per
ent of Ameri
an 
ompanies had a formal environmental poli
y statement or added environmental responsibility to 
ompany ethi
s statements by 1992 (Berenbeim (1992)). 21 2.4 Comparative stati
 analysis of an equilibrium where the most effe
tive VP is implemented In this se
tion, we examine how 
hanges in parameters affe
t the abatement rate of individual parti
ipating rms, the parti
ipation rates, and the aggregate abatement of the most effe
tive VP. For example, if an industry is more pollution-intensive, what happens to the parti
ipation rate, the redu
tion rate, and the aggregate abatement under a VP? Relative to the mandatory poli
y, how are the redu
tion rate and the aggregate abatement affe
ted by the pollution-intensity of the industry? What is the role of industry size, marginal 
osts, marginal damages, and politi
al dif
ulty in setting an effe
tive mandatory poli
y? The following proposition shows the effe
ts of 
hanges in these variables on the parti
ipation rate, the abatement rates of individual parti
ipating rms, and the aggregate abatement level. First, we examine the effe
ts of parameter 
hanges on the number of parti
ipating rms, abatement rates and aggregate abatement under the VP. Proposition 2.3. If the industry is more pollution-intensive, the size of the industry (the number of the polluting rms) is larger, or the slope of marginal damage is steeper, then the parti
ipation rate, redu
tion rate, and aggregate redu
tion under a VP in
rease. ( ¶ (N V P =N) ¶ ē i  0, ¶N V P a V ē i ¶ ē i  0, ¶a V ¶ ē i  0, ¶ (N V P =N) ¶N  0, ¶a V ¶N  0, ¶N V P a V ē i ¶N  0, ¶ (N V P =N) ¶d  0, ¶a V ¶d  0, and ¶N V P a V ē i ¶d  0.) Proof: See Appendix A.3. This result is quite intuitive (ex
ept for N). When the industry is pollution-intensive, the size of the industry is large, or if the slope of the marginal damage is steep, the so
ially optimal abate- ment rate and aggregate abatement are large. In su
h 
ases, the abatement rate and the aggregate abatement under the mandatory poli
y must also be high. Therefore, Proposition 3 implies that if the mandatory poli
y is more stringent, then the VP is also more stringent in the sense that the abatement rate, parti
ipation rate and aggregate abatement are higher. Relationships between pollu- tion intensity and the parti
ipation rate in the proposition are 
onsistent with the empiri
al ndings of the 33/50 program and the Canadian ARET program(e.g. Antweiler and Harrison (2007) and Gamper-Rabindran (2006)). 22 The impa
t of the in
rease in the size of the industry (the number of rms) on the parti
ipation rate may not be intuitive be
ause the larger the size of industry, the stronger the in
entive rms have to free-ride. However, lobby a
tivity and the VP suffer from free-riding. In addition, the so
ially optimal aggregate emission level does not 
hange even if the number of rms in
reases 9 . Be
ause the so
ially optimal aggregate emission level does not 
hange, the so
ially optimal aggregate abate- ment in
reases as mu
h as the aggregate emission level if the number of rms in
reases. Therefore, the so
ially optimal abatement rate of individual rms in
reases as the number of rms in
reases. Due to these two fa
tors, the mandatory standard, a kind of punishment for non-
ooperation for the VP, is mu
h more stringent under a larger industry, and the VP is also effe
tive. Be
ause an in
rease in the number of rms (and emissions per rm and slope of marginal damage) makes in
reased regulations desirable, it also makes the a
tual regulation level (abatement rate of the mandatory standard and VP and parti
ipation rate of the VP) higher. This is Proposition 2.3. It seems natural that the a
tual regulation level is higher if the desirable regulation level is higher. However, if the desirable regulation level is the same, how does the number of rms inuen
e the VP? Intuitively, due to free-riding, the VP for the small number of large polluting rms is more effe
tive than that for the large number of small rms in su
h a 
ase. However, the following Proposition shows that this intuition is wrong due to the free-riding on the VP. Proposition 2.4. The VP under a larger industry is more effe
tive than under a smaller industry if the aggregate natural emission levels are the same (and if the abatement 
ost, slope of marginal damage and politi
al dif
ulty in setting the mandatory poli
y are the same).(If Nē i = N 0 ē 0 i ; N > N 0 ; 
= 0 ; d = d 0 , and l = l 0 , then a V > a V 0 and N V P =N > N V 0 P =N 0 and N V P a V ē i > N V 0 P a V 0 ē 0 i .) Proof: See Appendix A.4. If the aggregate natural emission levels and other parameters (ex
ept for the pollution intensity of individual rms and the size of the industry) are the same, then the so
ially optimal aggregate abatement levels and rates are also the same. Proposition.4 implies that the VP for the larger industry is more effe
tive when the so
ially optimal aggregate abatement levels and rates are the 9 This is be
ause the marginal aggregate abatement 
ost is 
onstant and independent of the number of rms. 23 same. This result is 
ounterintuitive, but it indi
ates that the effe
tiveness of lobbying is eroded by free-riding. Remember a L = 1  dNē  
l dN 2 ē(1l ) . The third term ree
ts the effe
t of lobbying on the level of the mandatory standard, and it is greatly affe
ted by the size of the industry be
ause the third term is inversely proportional to N square. Thus, the industry size has signi
ant effe
ts on the effe
tiveness of lobbying when lobbying suffers from free-riding. The effe
tiveness of lobbying dire
tly determines the abatement rate of the VP and indire
tly deter- mines the parti
ipation rate of the VP or the seriousness of free-riding on the VP. First, we explain a me
hanism that determines the parti
ipation rate, then we explain how the me
hanism works when lobbying is ineffe
tive. The parti
ipation rate depends on the differen
e between the abate- ment rates of individual rms under the VP and the mandatory poli
y. If this differen
e is large (individual rms abate greatly under the VP relative to the mandatory poli
y), the parti
ipation rate is low be
ause the VP 
an generate a higher aggregate abatement than the mandatory poli
y even though the parti
ipation rate is low. If not, only a VP with a high parti
ipation rate 
an generate higher so
ial welfare than the mandatory poli
y. Thus, the differen
e between the abatement rates determines the parti
ipation rate of the VP or the degree of free-riding. The differen
e between the abatement rates of the VP and the mandatory poli
y is determined by the resour
es of individual rms reallo
ated from lobbying efforts to abatement under the VP. If the lobbying efforts are low, the differen
e between their abatement rates is small, and therefore, the parti
ipation rate is high a

ording to the dis
ussion in the last paragraph. Remember that the lobbying effort is small due to free-riding when the industry is large. Thus, we obtain results similar to Proposition 2.4 10 . We also examine the effe
ts of 
hange in parameters that lower the so
ially optimal abatement rate or the abatement rate under the mandatory standard. As we 
an guess from Proposition 3, the VP is also less stringent if the mandatory poli
y is less stringent. 10 However, if lobbying does not suffer from free-riding, the parti
ipation rate de
reases as the number of rms in
reases provided that aggregate emissions stay the same. In Appendix A.5, we analyze a 
ase where no rms free- ride on the lobbying of others. We formally show that under the assumption of Proposition 2.4, the VP is effe
tive and the parti
ipation rate is high if the industry is small (Proposition A.2). In 
onstrast with Proposition 4, Proposition 3 and the propositions shown below (Propositions 2.5 and 2.6) hold even though the lobbying a
tivity does not suffer from free-riding. 24 Proposition 2.5. If the industry's marginal 
osts in
rease or if the legislator is inuen
ed heavily by politi
al 
ontributions, then the parti
ipation rate, redu
tion rate, and aggregate redu
tion under a VP de
rease. ( ¶ (N V P =N) ¶  0, ¶a V ¶  0, ¶N V P a V ē i ¶  0, ¶ (N V P =N) ¶l  0, ¶a V ¶l  0, and ¶N V P a V ē i ¶l  0.) Proof: See Appendix A.3. Proposition 2.5 implies that the parti
ipation rate is likely to be low if l is high. A high l 
an be interpreted as high politi
al dif
ulty in setting the mandatory poli
y. Therefore, it may be good to implement the voluntary program with a low parti
ipation rate, as o

urred with the 33/50 program or the ARET program, if it is politi
ally dif
ult to set the mandatory standard when the program is initiated. The above two propositions explain the impa
t of 
hanges in parameters on the VP but do not state that the impa
ts on the VP are smaller or greater than they are on the mandatory poli
y. By 
omparing the impa
ts on the VP with those on the mandatory poli
y, we 
an evaluate their magnitude and better 
hara
terize VPs. The next proposition des
ribes the impa
t on VPs relative to that on the mandatory poli
y. Proposition 2.6. If a parameter 
hanges, the abatement rate of the VP will 
hange less than under a mandatory poli
y, but the aggregate abatement of the VP will 
hange more. For ē i , 0 ¶a V ¶ ē i  ¶a L ¶ ē i , and ¶N V P a V ē i ¶ ē i  ¶Na L ē i ¶ ē i  0 (The same relationships hold for N, and d). For 
, 0 ¶a V ¶  ¶a L ¶ and ¶N V P a V ē i ¶  ¶Na L ē i ¶  0 (The same relationships hold for l ). See Appendix A.3. Proposition 2.6 gives us two results. First, we explain why the redu
tion rate of the VP in
reases less than that of the mandatory poli
y. Changes in parameters affe
t politi
al 
ontributions. If politi
al 
ontributions de
rease, then a L in
reases. Although the total 
ost of rms under a manda- tory poli
y in
reases due to an in
rease in the abatement 
ost, the in
reased rate of the total 
ost is smaller than that of the abatement 
ost be
ause politi
al 
ontributions de
rease. This 
hange in total 
ost under the mandatory poli
y 
hanges the abatement rate of the VP less than that of the 25 mandatory poli
y be
ause the VP's abatement rate depends on the total 
ost under the mandatory poli
y. Proposition 2.6 also states that aggregate abatement under the VP is more sensitive to 
hanges in parameters than it is under the mandatory standard. Changes in parameters affe
t the aggregate abatement under the VP through two 
hannels: the number of parti
ipating rms, N V P , and the individual parti
ipating rms' abatement rate, a V . The effe
t of a V on the aggregate abatement under the VP is smaller than that of a L given the aggregate abatement under the mandatory stan- dard. However, due to the effe
t of N V P , 
hanges in parameters affe
t the aggregate abatement under the VP more than they do under the mandatory standard. More te
hni
ally, it should be re
alled that N V P = Na L =a V + 1, and therefore the aggregate abatement under the VP is a sum of the aggregate abatement under the mandatory standard and one rm's abatement under the VP (N V P a V ē i = Na L ē i +a V ē i ). The rst term is in
luded by (2.10), the 
onstraints on the number of parti
ipating rms 11 , and the se
ond term is determined by (2.8), the 
onstraints on the abatement 
ost or the abatement rate. We 
an divide the impa
t of the 
hanges in parameters on aggregate abatement into those that affe
t the 
onditions on the number of parti
ipating rms and those that affe
t the abatement rate. Be
ause the impa
t on the former 
ondition is the same as the impa
t on the aggregate abatement under the mandatory poli
y and the aggregate abatement under the VP is also inuen
ed by the latter, 
hanges in parameters inuen
e aggregate abatement under the VP more than they do under the mandatory poli
y. 2.5 Con
lusion We build a model with a regulator, a legislator and multiple polluting rms in whi
h the legislator, who is affe
ted by a lobby group representing polluting rms, sets the mandatory standard. We then used the model to explain why the regulator implements a VP. In this model, the regulator 
an implement the VP, whi
h generates lower so
ial 
osts than the mandatory standard, in 
ontrast to 11 (2.10) is not the dire
t 
onstraint on the number of parti
ipating rms but rather that on the aggregate abatement. However, only the number of parti
ipating rms is determined by (2.10) be
ause the abatement rate is determined by (2.8). 26 the model of Dawson and Segerson (2008). This differen
e in models o

urs be
ause we intro- du
ed an element of politi
al e
onomy into a mandatory poli
y-making pro
ess and the regulator in their model had a different obje
tive from the regulator in this 
hapter's model. The regulator's obje
tive in Dawson and Segerson (2008) was to a
hieve some aggregate emission level, but it was to minimize so
ial 
osts in this 
hapter. Be
ause the VP is subje
t to free-riding, the mandatory standard better a
hieves an aggregate abatement level than the VP does. Without politi
al inu- en
e, the regulator 
ould implement the mandatory poli
y, whi
h would a
hieve a so
ially-ef
ient out
ome. Thus, the differen
es between this 
hapter's model and Dawson and Segerson's shed important light on the ways a government should implement VPs. We found that the regulator should set the abatement rate under the VP at the highest possible level and should not set the abatement rate to maximize the parti
ipation rate. However, setting the redu
tion rate of the VP su
h that all polluting rms parti
ipate in the VP might be optimal when the legislator responds weakly or not at all to the lobbying of the polluting industries. Otherwise, setting the redu
tion rate at this level is not optimal. This study assumed that all rms are identi
al. However, one way to extend this 
hapter's model is to introdu
e the heterogeneity of polluting rms. Given that most voluntary approa
hes 
annot en- for
e a rm's 
ommitment, it would be interesting to 
onsider the 
ase in whi
h voluntary programs are not enfor
eable. Legislators have the right to make laws, and it is thus possible that legislators would set a mandatory standard even if a government or environmental organization de
ides to implement a VP. Exploring these types of extension remains an endeavor for future resear
h. 27 Chapter 3 Intra-Industry Spillover Effe
ts of ISO 14001 Adoption and Environmental Performan
e in Japan 3.1 Introdu
tion Voluntary approa
hes have be
ome in
reasingly popular approa
hes to environmental 
hallenges. One of the reasons for the popularity of voluntary approa
hes is that they are mu
h more exible than traditional 
ommand-and-
ontrol interventions. Voluntary approa
hes allow rms to redu
e emissions through more 
ost-effe
tive methods than the often less effe
tive mandated methods. Another reason for the popularity of voluntary approa
hes is that in
entive-based me
hanisms im- pose additional 
osts on rms beyond the expenses of pollution abatement, su
h as emission taxes or emission permit pur
hases. For these reasons, voluntary approa
hes are more a

eptable to rms than 
ommand-and-
ontrol or in
entive-based me
hanisms. Governments are promoting voluntary a
tions to address environmental issues for whi
h it is dif
ult to employ mandatory poli
ies. The introdu
tion of environmental management systems (EMSs) is one of the most 
ommon vol- untary a
tions performed by rms. In parti
ular, the ISO 14001 standard has re
eived in
reasing attention. A

ording to the International Organization for Standardization (ISO), a total of 223,149 
erti
ates had been issued worldwide by the end of 2009. In response to the popularity of ISO 14001, resear
hers have examined the reasons why fa
ilities adopt this standard. For example, by analyzing Japanese 
ompany data, Nakamura et al. (2001) and Nishitani (2009) nd that 
ertain 
hara
teristi
s of rms, su
h as their size, export ratio, debt ratio, pressures from stakeholders, and 28 their nan
ial exibility, affe
t their ISO 14001 adoption. Others also nd that rms are 
ertied to ISO 14001 earlier under greater regulatory pressures (King et al. (2005), Potoski and Prakash (2005a,b), and Darnall and Edwards (2006)). However, ISO 14001 has been 
riti
ized be
ause it does not expli
itly spe
ify an obje
tive or target for environmental performan
e. ISO 14001 is fo
used on operational pro
esses but not environ- mental out
omes. To obtain ISO 14001 
erti
ation, rms have to establish a Plan-Do-Che
k- A
tion (PDCA) 
y
le, whi
h is a 
ontinual 
y
le of planning (plan), implementing (Do), reviewing and improving the pro
esses (Che
k) and a
tions (A
tion) that are aimed at meeting the rms' own environmental targets and 
ontinually improving their environmental performan
e. Although ISO 14001 does not in
lude expli
it environmental performan
e obje
tives, it may nevertheless help rms improve their environmental performan
e. Therefore, the effe
tiveness of ISO 14001 adoption has also been examined in many studies. Some studies nd little eviden
e that ISO 14001 adoption has a positive effe
t on environmental performan
e (Barla (2007), King et al. (2005), and Darnall and Side (2008)), whereas a number of other studies nd substantial eviden
e of that effe
t (Arimura et al. (2008), Potoski and Prakash (2005a,b), and Melnyk et al. (2003)). A key 
ontri- bution of this 
hapter is to examine whether ISO 14001 has made a positive 
ontribution to the environmental performan
e of Japanese rms. Spillovers between rms were not 
onsidered in previous studies on the adoption and effe
tiveness of ISO 14001, ex
ept for Arimura et al. (2009), who estimates the effe
ts of ISO adoption on green supply 
hain management. If spillover effe
ts exist among rms within the same industry, estima- tion results of the effe
t of ISO 14001 on environmental performan
e will be biased. A

ording to an international survey 
ondu
ted by the Organization for E
onomi
 Co-operation and Devel- opment (OECD) in Canada, Fran
e, Germany, Hungary, Japan, Norway, and the US, the fa
t that similar fa
ilities are adopting EMS and similar environmental pra
ti
es motivates other rms to adopt them. Thus, the de
isions by rms and fa
ilities on ISO 14001 adoption and environmental performan
e are likely to be inuen
ed by those of other rms and fa
ilities within the same in- dustry. If su
h externalities exist, it is important not only to 
ontrol for them but also to measure their magnitudes when we estimate the effe
t of 
ontributing fa
tors on ISO 14001 adoption and environmental performan
e. In addition, if intra-industry spillovers exist, governments 
an pursue 29 industry-spe
i
 voluntary programs in 
ooperation with industry asso
iations. Employing an extensive dataset of Japanese produ
tion fa
ilities a
ross a large number of indus- tries, this 
hapter examines the determinants of ISO 14001 adoption, the effe
t of ISO 14001 adop- tion on environmental performan
e, and the existen
e of intra-industry spillovers. To 
ontrol for and estimate the spillovers, we employ a Bayesian spatial autoregressive (SAR) probit model for ISO 14001 adoption and a SAR model for two types of emission redu
tions (emissions into air and emissions into water). The potential for intra-industry spillovers is based on the hypothe- sis that fa
ilities that adopt ISO 14001 (or redu
e emissions) do so be
ause other fa
ilities in the same industry also adopt ISO 14001 (or redu
e emissions). This type of spillover has been ex- amined by a few empiri
al studies on voluntary environmental approa
hes (in
luding ISO 14001). Only Antweiler and Harrison (2007) 
onsiders spillovers, and their results show an intra-industry spillover of parti
ipation in the Canadian voluntary ARET (A

elerated Redu
tion/Elimination of Toxi
s) program. However, they do not estimate the effe
t of spillovers on emission redu
tions. By employing spatial e
onometri
 estimation methods, we nd that there are positive spillovers of ISO 14001 adoption for redu
ing water emissions between Japanese fa
ilities that belong to rms with similar revenue levels in the same industry. Plants that adopt ISO 14001 are more likely to do so if their industry peers also adopt ISO 14001. In addition, the per
entage 
hange in the weighted sum of emissions into the air, used as a measure of environmental performan
e, is 
orrelated between similar sized fa
ilities in the same industry. This 
orrelation might ree
t the fa
t that similar sized fa
ilities in the same industry have similar te
hnologies and therefore have similar environmental performan
e. There are three other noteworthy results. First, plant size (as determined by the number of workers in a fa
ility) has a signi
antly positive effe
t on ISO 14001 adoption, 
onrming the results from previous studies on ISO 14001 adoption. Se
ond, fa
ilities that emit into water with lower water emission intensities 
ompared to other fa
ilities are more likely to adopt ISO 14001. Third, there is no statisti
ally signi
ant eviden
e that ISO 14001 adoption improves environmental performan
e. This 
hapter is organized as follows. Se
tion 2 briey explains the ba
kground of this resear
h, the ISO 14001 standard, and the reasons for employing a Japanese dataset for this study. This se
tion 30 also dis
usses the rationale for examining intra-industry spillovers. We explain our e
onometri models in Se
tion 3, and in Se
tion 4, we introdu
e the data from the empiri
al analysis of this 
hapter. Se
tion 5 shows estimation results, and then we 
on
lude the paper in Se
tion 6. 3.2 Ba
kground and hypotheses on spillover effe
ts 3.2.1 ISO 14001 in Japan In this se
tion, we explain the reasons for employing a Japanese dataset to examine the determinants of ISO 14001 adoption and its effe
t on environmental performan
e and why we have to examine intra-industry spillover effe
ts. ISO 14001 is an internationally re
ognized standard for an environmental management system (EMS) that was released in 1996 and revised in 2004 by the International Organization for Stan- dardization. Firms 
an adopt ISO 14001 at the level of the individual fa
ility, group(s) of fa
ilities, or the entire 
ompany. To adopt ISO 14001, fa
ilities are 
ertied by external third-party reg- istrars. Certied fa
ilities must follow a 
y
le of Plan-Do-Che
k-A
t over time: environmental planning (“Plan”), plan implementation and operation (“Do”), monitoring (“Che
k”), 
orre
tive a
tion (“A
tion”), and management review. As shown in Table 3.1, ISO 14001 has been very popular in Japan sin
e its release. ISO 14001 adoption has in
reased more rapidly in Japan than in other 
ountries ex
ept China after 2003. A
tually, Japan's share of the total number of ISO 14001 
erti
ates was more than 20% until 2005. Although the total number of 
erti
ates in China (39,195) ex
eeded that of Japan (35,573) in 2008, this number is more than twi
e that of Spain, whi
h had the third highest number of 
ertied organizations (16,443). Due to the popularity of ISO 14001 in Japan, many studies on ISO 14001 employ Japanese data at the rm or fa
ility level (e. g. , Arimura et al. (2008), Arimura et al. (2009), Nakamura et al. (2001), Nishitani (2009), and Wel
h et al. (2002)). The high adoption rate of ISO 14001 a
ross a large number of industries in Japan is a prime reason for employing su
h data in this study as well. 31 Table 3.1: Top 3 for ISO 14001 
erti
ates in 2000, 2003, and 2008 2000 2003 2008 Japan 5556 Japan 13416 China 39195 UK 2534 UK 5460 Japan 35573 Sweden 1370 China 5064 Italy 16443 Previous studies of Japanese fa
ilities or rms found that fa
ility/rm size has a signi
ant positive impa
t on the adoption of ISO 14001 environmental management standards (see Arimura et al. (2008), Arimura et al. (2009), and Wel
h et al. (2002) for fa
ility level analysis and Nakamura et al. (2001) and Nishitani (2009) for rm level analysis). This relationship likely exists be
ause the 
osts of ISO 14001 adoption are less signi
ant for large fa
ilities or rms than for small ones. In addition, foreign 
ustomers who nd it dif
ult to monitor the performan
e of overseas rms may require ISO 14001 adoption as a visible sign of 
ommitment to environmental prote
tion. Therefore, fa
ilities or rms with more foreign 
ustomers are more likely to adopt ISO 14001; previous studies have found eviden
e to 
onrm this 
on
lusion (Arimura et al. (2009), Arimura et al. (2008), Nakamura et al. (2001) and Nishitani (2009)). ISO 14001 adoption in Japan might be stimulated by governmental poli
ies. Many lo
al govern- ments have en
ouraged the adoption of ISO 14001 through nan
ial support and/or informational support, su
h as seminars on ISO 14001. Fa
ilities in our sample were lo
ated in 1,056 muni
- ipalities, and 105 of them provided some support for ISO 14001 adoption in 2001. Some lo
al governments have also adopted ISO 14001 and have provided advi
e on ISO 14001 adoption based on their own experien
e. However, in re
ent years, Japanese governments have en
ouraged EMS 
erti
ations other than ISO 14001 be
ause most large rms adopted ISO 14001 around 2005, while it remained too ex- pensive for small and medium-sized enterprises (SMEs) to adopt ISO 14001. Therefore, Japanese governments started en
ouraging SMEs to adopt EMSs that are less expensive than ISO 14001, su
h as E
o-A
tion 21, whi
h was laun
hed by the Japanese Ministry of the Environment in 1996 and be
ame an EMS with a third-party 
erti
ation in 2004. Be
ause 
omplete information is not available on the registration of 
ertain EMSs that target SMEs, we employ emission data during 32 2001-2003, when there was a small number of registrations of these alternative EMSs. The EMS adoption de
ision by fa
ilities likely inuen
es the EMS adoption de
isions of other rms in the same industry. An OECD survey, “Environmental Poli
y Tools and Firm-Level Management and Pra
ti
es in Japan, ” shows that 53% of Japanese fa
ilities 
onsider what “other fa
ilities like ours are adopting” as “very important” or “important” as motivation for adopting an EMS(Hibiki and Arimura (2004)). Therefore, it is very likely that spillover effe
ts of ISO 14001 adoption exist within an industry. The above OECD survey was 
ondu
ted not only in Japan but also in Canada, Fran
e, Germany, Hungary, Norway, and the US. This multi-
ountry survey also asked rms about their motivations to implement environmental pra
ti
es. Spe
i
ally, rms were asked about the importan
e of the fa
t that similar fa
ilities were adopting similar environmental management pra
- ti
es 1 . From 20% to 60% of rms 
onsider that this fa
t is moderately important or very important (Darnall and Pavli
hev (2004), Gla
hant et al. (2004), Kerekes et al. (2004) Rennings et al. (2004), and Ytterhus (2004)). It is very likely that some, or even many, Japanese rms think and a
t simi- larly. Therefore, both the de
ision to adopt ISO 14001 and the de
ision to improve environmental performan
e may be inuen
ed by the a
tions of other rms (or their fa
ilities). 3.2.2 Hypotheses on spillovers How should we 
onstru
t weight matri
es to 
apture spillover effe
ts? The answer depends on what types of spillover effe
ts we want to examine. In this subse
tion, we dis
uss hypotheses about the effe
ts of intra-industry spillover on the de
ision to adopt ISO 14001 and on environmental performan
e. As mentioned in the previous subse
tion, the OECD survey revealed that many Japanese fa
ilities were motivated to adopt ISO 14001 by the knowledge that other fa
ilities/rms in the same in- dustry had adopted the standards. If we interpret this evaluation literally, industry-wide spillover effe
ts likely exist for the adoption of ISO 14001: fa
ilities adopt ISO 14001 be
ause other fa
ili- ties in the same industry have adopted it. In addition, the OECD survey showed that many fa
ilities 1 This question was posed to rms in Canada, Germany, Hungary, Norway, and the US but unfortunately not to rms in Japan. 33 
onsidered the environmental pra
ti
es adopted by similar fa
ilities to be an important motivation for the adoption of similar pra
ti
es. If fa
ilities with similar levels of 
apital intensity and te
h- nology adopt similar environmental pra
ti
es, their environmental performan
e is likely 
orrelated and may ree
t intra-industry 
orrelation or spillover effe
ts. Therefore, we examine three main hypotheses. Hypothesis 1 There are industry-wide spillover effe
ts or 
orrelations that affe
t ISO 14001 adop- tion and environmental performan
e. As dis
ussed above, the OECD survey indi
ates that fa
ilities may imitate or adopt the same pra
- ti
es as other fa
ilities. Su
h imitation 
an o

ur for 
ompetitive reasons. For example, rms are likely to adopt the same pra
ti
e if they think that not doing so would redu
e their 
ompetitiveness in the marketpla
e. Su
h imitation pressure for 
ompetitive reasons is likely to be strong when the pra
ti
e is widespread. The adoption of ISO 14001 was already widespread in Japan in 2000, as seen in Table 3.1. Imitation pressure be
ause of 
ompetition may therefore 
ontribute to the diffusion of ISO 14001. A fa
ility/rm is likely to have a greater inuen
e on the de
isions and a
tions of similar fa
ili- ties/rms than on those with different 
hara
teristi
s. This is espe
ially true if the most signi
ant determinants of the de
isions and a
tions of fa
ilities differ; those fa
ilities will have little inuen
e on ea
h other and will a
t more independently. For example, many studies, su
h as Nakamura et al. (2001), Wel
h et al. (2002), Arimura et al. (2008) and Nishitani (2009), have found that fa
ilities with more workers are more likely to adopt ISO 14001; this is likely true be
ause of the 
ompara- tively high initial 
ost of adoption. This nding implies that large fa
ilities have a strong in
entive to adopt ISO 14001, whereas smaller fa
ilities may not have an in
entive to adopt it. The adoption of ISO 14001 by large fa
ilities may not affe
t the de
isions of small fa
ilities. There may also be a negative 
orrelation between ISO 14001 adoption at large and small fa
ilities be
ause they are likely to have very different adoption in
entives. Therefore, we 
onsider spillover effe
ts and 
orrelations among similar fa
ilities. In parti
ular, we fo
us on two types of similarities. First, we fo
us on fa
ility size as measured by the number of workers. As mentioned above, fa
ility size is a signi
ant determinant of ISO 14001 adoption. This 34 makes logi
al sense be
ause fa
ilities that have more workers are likely to be able to allo
ate more workers to ISO 14001 adoption. Thus, fa
ilities with a similar number of workers may have similar attitudes toward ISO 14001 adoption and therefore make similar de
isions. Thus, intra-industry 
orrelations or spillover effe
ts may exist for ISO 14001 adoption at similarly sized fa
ilities. In addition, there may be intra-industry 
orrelations or spillover effe
ts for environmental perfor- man
e at similarly sized fa
ilities. Similarly sized fa
ilities in the same industry are likely to have similar levels of 
apital intensity and te
hnologies, and they are therefore likely to have similar environmental performan
e. Hypothesis 2 There are intra-industry spillover effe
ts or 
orrelations between similarly sized fa- 
ilities that affe
t ISO 14001 adoption and environmental performan
e. The se
ond type of similarity we fo
us on is the size of the rms to whi
h fa
ilities belong. Simi- larly sized rms in the same industry are likely to be 
ompetitors and therefore may adopt similar environmental pra
ti
es. Among the variables in our dataset, revenue is most able to 
apture rm size or demonstrate “rivalry". Therefore, fa
ilities that belong to rms with similar levels of revenue may inuen
e ea
h other's de
isions and a
tions on environmental issues. Hypothesis 3 There are intra-industry spillover effe
ts or 
orrelations among fa
ilities that belong to similarly sized rms (as determined by revenue) that affe
t ISO 14001 adoption and envi- ronmental performan
e. Intra-industry spillover effe
ts among similarly sized rms (as determined by revenue) is likely to explain the ISO 14001 adoption de
isions of fa
ilities better than the spillover effe
ts among similarly sized fa
ilities. Be
ause of the initial 
ost, the ISO 14001 adoption de
isions of many fa
ilities are not made by the fa
ilities themselves but by the rms to whi
h they belong. However, spillover effe
ts among similarly sized fa
ilities are likely to explain environmental performan
e better than the other hypotheses be
ause fa
ility size may be a more important determinant of environmental performan
e. In the next se
tion, we will dis
uss our strategy for evaluating these three hypotheses. 35 3.3 Estimation strategy The models that we employ are essentially spatial e
onometri
 models. For emission redu
tion, we employ a spatial autoregressive (SAR) model. For ISO 14001 adoption, we employ a SAR probit model. We will not dis
uss the estimation methods in detail here; LeSage and Pa
e (2009) for details (in parti
ular, Chapters 3, 5, and 10). Let E i jklt and DE i jklt (= E i jklt+1 =E i jklt ) be the emission level of fa
ility i of 
ompany j in industry k at muni
ipality l and the emission redu
tion (
hange) of fa
ility i from year t to year t+1, respe
- tively. We normalize both E i jklt by fa
ility size (the number of workers in the fa
ility) and denote by EL i jklt this normalized emission level. We 
an interpret EL i jklt as the proxy of emission inten- sity in terms of fa
ility size be
ause the fa
ility size is very likely to be 
losely 
orrelated with the number of workers. We employ EL i jklt as the proxy of emission intensity due to data availability issues although it might be better to dene emission intensity dividing by output or revenue. We will dis
uss data availability in the next se
tion. We assume that the emission redu
tion equations take the following form. DE i jklt = r å m w im DE mjklt +b F X F it +b C X C jt +b M X M lt + g ˆ ISO i jklt + e i jklt (3.1) or DE t = rWDE t +b F X F t +b C X C t +b M X M t + g ˆ ISO t + e t : (3.2) X F it , X C jt , and X M lt are 
hara
teristi
s of fa
ility i, rm j, and muni
ipality l, whi
h are explained below.W is a (standardized) weight matrix whose elements are w i j = 1=m i if fa
ility j is one of the m i nearest neighbors of i (but w ii = 0), and w i j = 0 otherwise. We will dis
uss the sele
tion 
riteria for the nearest neighbors at the end of this se
tion. To 
ontrol for the endogeneity of ISO 14001 adoption, ˆ ISO i jklt is the ISO 14001 adoption modeled using a latent variable, ISO  i jklt . Con
retely, ˆ ISO i jklt = 1 if ISO  i jklt > 0 and ˆ ISO i jklt = 0 otherwise. The ISO 14001 adoption latent variables are 36 estimated by ISO  t = nWISO  t +k F X F t +k C X C t +k M X M t +lEL t +m t (3.3) m t  N(0; I n ): In the next subse
tions, we briey explain our methods and pro
edures for estimating (3.2) and (3.3). 3.3.1 Estimation of environmental performan
e equation We estimate (3.2) byMaximum likelihood (ML). Con
retely, we assume e t N(0; I n ) and minimize the following log-likelihood fun
tion min 1<r<1 logL(r) =(n=2) log(p)+ log j I n rW j (n=2) log(S(r)) where S(r)= (e o re d ) 0 (e o re d ); e o = yX ˆ b o ; e d = yX ˆ b d ; ˆ b o =(X 0 X) 1 X 0 y; ˆ b d =(X 0 X) 1 X 0 Wy, and X = (X F ;X C ;X M ). Note that 1 < r < 1 must hold for the minimization problem to be solvable. After solving this optimization problem and determining optimum ˆ r , we then derive ˆ b = ( ˆ b 0 F ; ˆ b 0 C ; ˆ b 0 M ) 0 from ˆ b = ˆ b o r ˆ b e . 3.3.2 Estimation of the ISO 14001 adoption equation We estimate the equation (3.3) by MCMC sampling whi
h is a standard method for estimating spatial probit models. We need 
onditional distributions in order to sample the latent variable ISO  i jklt and need prior distribution(s) (n;k F ;k C ;k M ;l ) to derive the 
onditional 
ontributions. We assume prior independen
e between n , and q , where q = (k F ;k C ;k M ;l ). We also assume a univariate uniform pdfU(1;1) for n , and a multivariate-normal pdf N(
;T ) for q . Let the observed ISO 14001 adoption, ISO it , be equal to 1 if fa
ility i adopts ISO 14001 at year t 37 and equal to 0 otherwise. Then, the joint posterior distribution is given by p(q ;njISO it )µ jAjexp( 1 2 m 0 t m t )p(q)p(n) (3.4) where A= I n nW and p(q) and p(n) are prior on q and n . The 
onditional distributions are p(q jn; ISO t ; ISO  t ) µ exp((AISO  t Xq) 0 (AISO  t Xq))exp((q  
) 0 T 1 (q  
)) (3.5) p(njq ; ISO t ; ISO  t ) µ exp((AISO  t Xq) 0 (AISO  t Xq)) (3.6) p(ISO  t jq ;n; ISO t ) µ exp((AISO  t Xq) 0 (AISO  t Xq)) (3.7) where X t = (X F t X C t X M t E t ). (3.5) implies that q is distributed f MVN with mean M and varian
e V where M = (X 0 X+T 1 ) 1 (X 0 AISO  t +T 1 
) and V = (X 0 X+T 1 ) 1 . We sample the n using a random-walk Metropolis-Hastings pro
edure be
ause they do not 
orre- spond to any known probability distribution. Candidate variables n C are produ
ed by using stan- dard normal distributions and tuning parameters d as follows: n C = n  +dN(0;1) Then, we a

ept these 
andidates by probability, p=min  1; f (n C jq ; ISO t ; ISO  t ) f (n  jq ; ISO t ; ISO  t )  ; (3.8) where f (n  jq ; ISO t ; ISO  t ) = ((AISO  t Xq) 0 (AISO  t Xq)): (3.9) By following LeSage and Pa
e (2009), we adjust tuning parameters based on monitoring the a

ep- tan
e rates from the Metropolis-Hastings pro
edure during the MCMC sampling. If the a

eptan
e rate is smaller than 40%, then we update d 0 = d=1:1. If the a

eptan
e rate is greater than 60%, then d 0 = 1:1d. 38 (3.7) is proportional to themultivariate normal distributionwithmean A 1 Xq(= k=(k 1 ;k 2 ;    ;k n ) 0 )and varian
e (A 0 A) 1 subje
t to trun
ation 
onstraints, whi
h depends on the observed value 0 or 1 for ISO. We sample the individual element ISO  it using Geweke's (1991) approa
h. We sample z i 
onditional on z i under the trun
ation 
onstraints and 
onstru
t ISO  it = k i + z i . Let Y = A 0 Aand g i = Y i;i =Y ii where Y i;i is theith row of Y without the ith element. Then, E(z i jz i ) = g i z i , and therefore, a normal 
onditional distribution for z i is obtained as follows; z i jz i = g i z i +(Y ii ) 1=2 v i (3.10) where v i  N(0;1) under the trun
ation 
onstraints, v i < (k i  g i z i )(Y ii ) 1=2 (, ISO  i < 0) i f ISO i = 0 v i > (k i  g i z i )(Y ii ) 1=2 (, ISO  i > 0) i f ISO i = 1: We sample v i from N(0;1) under the trun
ation 
onstraints. We implement m-step Gibbs sampling to produ
e z. On the initial step, we set z to zeros, and at the jth iteration of z i , we sample z j i using z j 1 ;z j 2 ; :::;z j i1 and z j1 i+1 ;z j1 i+2 ; :::;z j1 n where z j i is the value after the j iteration. We 
ontinue this pro
edure for m iterations for all i. Our sampling routine for estimating the parameters is 1. determine q by drawing from f MVN (M;V ); 2. determine n by generating 
andidate n using a random-walk pro
edure and a

epting them with probability p in (3.8); 3. determine ISO  by generating z m by sampling v i and updating z j i for i = 1;2; ::;n and j = 1;2; :::;m as explained above, and setting ISO  = k+ z m . 3.3.3 Weight matri
es We employ three types of weight matri
es to 
apture and examine the intra-industry spillover effe
ts dis
ussed in Se
tion 2. The rst weight matrix is intended to 
apture and examine spillover effe
ts 39 between fa
ilities in the same industry. Be
ause fa
ilities are likely to be more inuen
ed by other fa
ilities with similar 
hara
teristi
s, the se
ond and third matri
es are formed to 
apture and examine spillover effe
ts between “similar” fa
ilities within industries, in terms of the number of workers and revenue levels of the rms. For 
omparison, we also estimate models with/without spatial 
orrelations with all other fa
ilities. More formally, we estimate models with the following ve weight matri
es. 1. Weight matrixW = 0 (no spatial 
orrelation) (Weight Matrix I (WM I)) 2. For ea
h i,W ii = 0 andW i j = 1=(n1) 8 j 6= i. (
orrelated with all other fa
ilities) (WM II) 3. For ea
h i,W i j = 1=m i 8 j 6= i if fa
ility iand j are in the same industry. (
orrelated with all other fa
ilities in the same industry, m i is the number of fa
ilities in the same industry as i) (WM III) 4. For ea
h i,W i j = 1=m for any fa
ility j ( j 6= i ) that is in the same industry as i and one of the m nearest fa
ilities in terms of the number of workers in the fa
ility 2 . (WM IV) 5. For ea
h i, W i j = 1=m for fa
ility j ( j 6= i ) that is in the same industry as i (but belongs to different rms) and one of the m nearest fa
ilities in terms of revenue 3 . (WM V) For WM IV and V, we employ m = 3;5;and10. WM III is 
onstru
ted to examine Hypothesis 1, WM IV for Hypothesis 2, and WM V for Hypothesis 3. If some element W i j of W is positive, the de
isions or a
tions of fa
ilities i and j are assumed to 
orrelate with ea
h other or inuen
e ea
h other. Therefore, if W i j is positive, we 
an examine whether there is a spillover effe
t or 
orrelation in the de
isions and a
tions of fa
ilities i and j. However, W i j = 0 implies that we assume that there is no 
orrelation or spillover effe
t in the de
isions or a
tions of fa
ilities i and j. Thus, theWM I model is similar to a standard probit model 4 be
ause all elements of W are equal to 0, and there are no 
orrelations between the de
isions or a
tions of various fa
ilities. 2 If several fa
ilities have equal numbers of employees, we 
onsider the fa
ility that belongs to a rm with more similar revenue as nearer. 3 If several fa
ilities belong to rms with equal revenue, we 
onsider the fa
ility with the more similar number of workers to be nearer. 4 There is a slight differen
e be
ause we employ a Bayesian estimation method to estimate this model. 40 In other models, spillover effe
ts between the de
isions or a
tions of some fa
ilities are thought to exist. In the WM III model, we assume that there are 
orrelations between the de
isions or a
tions of all fa
ilities in the same industry but no 
orrelation between any fa
ilities in different industries. Thus, we use this model to examine intra-industry and industry-wide spillovers, i.e., Hypothesis 3. The nearest fa
ility in terms of number of workers is the fa
ility of most similar size, whereas the nearest fa
ility in terms of revenue is equal to the fa
ility that belongs to the most similarly sized rm (in terms of revenue). Therefore, theWM IVmodel 
aptures the intra-industry spillover effe
ts between similar sized fa
ilities, and the WMV 
aptures the intra-industry spillover effe
ts between fa
ilities that belong to similar sized rms. If m is smaller, then we examine spillover effe
ts within narrower limits (between the larger numbers of fa
ilities). The WM II model 
an 
apture inter-industry spillover effe
ts. However, we 
onstru
t this model mainly to 
he
k whether the intra-industry spillover effe
ts of the three models above are signi
ant and are not due to unobservable effe
ts on all fa
ilities. In other words, this model is built to examine whether those three models pre
isely 
apture the intra-industry spillover effe
ts that we want to examine and whether the intra-industry spillover effe
ts have explanatory power for ISO 14001 adoption and environmental performan
e. 3.3.4 Model 
hoi
e We employ models with 5 types of weight matri
es for both the rst and se
ond stages. How- ever, we 
onstru
t ˆ ISO i jklt by the rst stage model whi
h is supported by the Bayesian or Akaike Information Criterion a

ording to Hepple (2004). 3.4 Data des
ription 3.4.1 Emission data We 
olle
t data on the emission and transfer of 354 
hemi
al substan
es from the PRTR webpage. These data are available from 2001. We 
onsider emissions into air and water and 
al
ulate the 41 toxi
ity-weighted sum of all dire
t emissions into air and water from ea
h fa
ility (E A t+1 =E A t and E W t+1 =E W t ), respe
tively. For this 
al
ulation, we employ the US Environmental Prote
tion Agen
y (2010) Risk-S
reening Environmental Indi
ators (RSEI) whi
h are also employed by Antweiler and Harrison (2007) and Potoski and Prakash (2005a). We 
onstru
t the per
entage 
hanges in the weighted sum of emissions into air and water (E A t+1 =E A t and E W t+1 =E W t )and use them as environ- mental performan
e indi
ators be
ause many rms set their emission redu
tion goals in terms of per
entage redu
tion relative to the previous year. 3.4.2 ISO 14001 adoption data We 
an 
olle
t data on ISO 14001 adoption and the time of adoption for fa
ilities that were 
ertied by Japanese 
erti
ation bodies from the Japan A

reditation Board for Conformity Assessment (JAB). Data on ISO 14001 
erti
ations by foreign 
erti
ation bodies are 
olle
ted from the rms' websites. 3.4.3 Data on other variables Data on other 
hara
teristi
s of rms, su
h as their revenue, prot, and age, are 
olle
ted from Japanese 
ompany data books (Teikoku Data Bank (2003a,b, 2004a,b, 2005a,b)). Pressure from residents near a fa
ility 
an motivate a 
ompany to implement voluntary a
tions on environmental issues, and the in
ome and population density of the lo
al 
ommunity are very likely determinants of the level of pressure for a greener environment. Therefore, we 
olle
ted data on muni
ipal 
har- a
teristi
s, su
h as in
ome, population, area (for 
al
ulation of population density) from the CD of the Japanese“System of So
ial and Demographi
 Statisti
s”. In addition, the author obtained data on the number of grievan
es against environmental pollution for ea
h muni
ipality from the Environmental Dispute Coordination Commission to dis
lose and determined whether the muni
i- palities provided informational and nan
ial support for ISO 14001 adoption. Support from most of the muni
ipalities targeted small and medium-sized enterprises (SMEs), but not all of the SMEs that adopt(ed) ISO 14001 re
eive(d) the support. 42 From the PRTR database, we determine the number of employees in ea
h fa
ility and 
onstru
t the emissions per employee as a measure of emission intensity. To 
onstru
t the emission intensity, output-based measures, su
h as revenue or prot, might be more ideal than the number of employ- ees. However, revenue and prot are 
omputed at the rm level rather than at the fa
ility level. In our sample, 252 of 663 fa
ilities that emit into water (38.0%) belong to multi-fa
ility rms, as do 1,793 of 3,579 fa
ilities that emit into air (50.1%). Therefore, rm level variables are not suitable for 
al
ulating the emission intensity at the fa
ility level, and therefore, we employ the emissions intensity in terms of workers in the fa
ility. Using employment as a proxy for size might lead to distortions of emission intensity be
ause 
apital intensity varies a
ross industries. It is reasonable to assume that 
apital intensities are relatively similar within a given industry. Therefore, emis- sion intensities normalized to employment are 
omparable within that industry. For 
omparisons a
ross industries, the industry dummies or xed effe
ts will a

ount for the differen
es in 
apital intensities. 3.4.4 Des
riptive statisti
s Tables 3.2 and 3.4 show the des
riptive statisti
s of the main variables (in 2001) used for the esti- mation. Be
ause we estimate the determinants of ISO 14001 adoption and its effe
t on redu
tion of emissions into air and water separately, we show the des
riptive statisti
s of fa
ilities that emit into water (Table 3.2) and air (Table 3.4) separately 5 . Note that the average air emission inten- sity de
reased from the year 2001 to 2003, while the average water emission intensity in
reased during that period. On average, fa
ilities that emit into air were lo
ated in muni
ipalities with higher population densities, wealthier residents, and more grievan
es against environmental pollu- tion 
ompared to fa
ilities that emit into water. In addition, a higher per
entage of fa
ilities that pollute the air (22.1%) were lo
ated in muni
ipalities that provide informational or nan
ial sup- port for ISO 14001. Fa
ilities that emit into water had more workers and more intense emissions on average. Tables 3.3 and 3.5 show the industry 
omposition and ea
h industry's average of the main variables 5 505 fa
ilities emit into both air and water, so the total number of fa
ilities in the two datasets is 3737. 43 Table 3.2: Des
riptive statisti
s of water emitting fa
ilities in 2001(N=663) mean s. d. max min Workers/100 6.88 13.5 190 0.01 Weighted sum of emissions to water/10 6 (E W ) 25.0 360 8860 6.30E-07 E W in 2003 28.4 360 6560 0 Air emission intensity (kg*weight/persons)/10 6 0.0251 0.153 2.75 0 Water emission intensity (kg*weight/persons)/10 6 0.0809 1.25 31.8 9.84E-10 Prot/revenue (yen/thousand yen)/1000 -6.87E-04 0.0957 1.73 -0.959 Population density (persons/ha)/100 0.168 0.210 1.21 0.00140 In
ome per 
apita (thousand yen/persons)/1000 1.44 0.233 2.76 0.423 Grievan
es against environmental pollution/1000 0.0736 0.120 0.783 0 ISO 14001 0.585 0.493 Support 0.198 0.398 for water and air emitting fa
ilities. In the water emitting fa
ility sample, there are 11 industries, and many of these fa
ilities also emit into the air. Equipment and supplies are the three largest. However, there are 21 industries in air emitting fa
ility sample, and relatively fewer of these fa
ili- ties also emit into water. 3.5 Estimation results The estimation results of fa
ilities that emit into water and air are presented separately. In the sampling routine of the SAR probit models, we utilize 200 draws for the burn-in period and then 1000 draws for the a
tual sampling with a 3-step Gibbs sampling of z; we set 
 = (0; :::;0) 0 , T = I k  1:00e+ 05, and d = 0:1(the initial value) where k is the number of explanatory variables. We use a relatively faster desktop 
omputer (with CPU Core i7 and 8 GB of memory) and R for estimation, and approximately 0.5-5 hours and 9-18 hours are required to estimate the SAR probit models with 1326 and 7182 observations, respe
tively. For the WM IV and V models, we show only the models with 'm' that have the greatest likeli- hood(the greatest likelihood is equal to the greatest AIC and BIC). For other `m' values, please see Table B.1-B.4 in the Appendix B. 44 Table 3.3: Industry 
omposition of water emitting fa
ilities with an average of main variables Name of industry Num of fa
i. ISO 14001 E W Fa
i. w/ E A = 0 Workers Textile mill produ
ts Mfg. 25 40.0% 1.02 9 309.92 Pulp, paper and paper produ
ts Mfg. 27 66.7% 1.15 3 467.78 Chemi
al and allied produ
ts Mfg. 216 50.5% 62.6 32 294.00 Plasti
 produ
ts Mfg. 20 55.0% 9.17 1 430.10 Cerami
, stone and 
lay produ
ts Mfg. 36 58.3% 0.396 3 451.30 Iron and steel Mfg. 31 58.1% 29.9 5 1005.90 Non-ferrous metals and produ
ts Mfg. 36 33.3% 20.7 9 433.39 Fabri
ated metal produ
ts Mfg. 100 36.0% 6.32 54 205.73 Ele
tri
al ma
hinery, equipment and supplies Mfg. 96 84.4% 1.63 24 1230.24 Transportation equipment Mfg. 66 69.7% 5.21 4 2410.46 Mis
ellaneous manufa
turing industries 10 40.0% 0.268 5 254.80 Table 3.4: Des
riptive statisti
s of air emitting fa
ilities in 2001 (N=3579) mean s. d. max min Workers/100 3.63 8.11 190 0.01 Weighted sum of emissions to air/10 6 (E A ) 2.67 17.6 362 1.55E-06 E A in 2003 1.84 12.2 396 0 Air emission intensity (kg*weight/persons)/10 6 0.0156 0.145 6.23 1.53E-09 Water emission intensity (kg*weight/persons)/10 6 0.00449 0.0848 3.53 0 Prot/revenue (yen/thousand yen)/1000 0.00382 0.0798 1.73 -2.64 Population density (persons/ha)/100 0.226 0.290 1.70 0.00113 In
ome per 
apita (thousand yen/persons)/1000 1.44 0.233 2.76 0.423 Grievan
es against environmental pollution/1000 0.0821 0.130 0.783 0 ISO 14001 0.414 0.493 Support 0.221 0.415 3.5.1 ISO 14001 adoption and performan
e of fa
ilities that emit into water Table 3.6 shows the estimation results on ISO 14001 adoption by fa
ilities that emit into water. The number of workers in a fa
ility has a signi
antly positive effe
t on ISO 14001 adoption at the 1% level under all spe
i
ations, whereas the water emission intensity has a negative effe
t at the 5% or 10% level. However, support for ISO 14001 adoption, protability, and the 
hara
teristi
s of the muni
ipality in whi
h a fa
ility is lo
ated do not have signi
ant effe
ts. The impli
ations of these estimation results are as follows. First, a fa
ility with more workers is more likely to adopt ISO 14001; these ndings are similar to those of Nakamura et al. (2001), Wel
h et al. (2002), Arimura et al. (2008) and Nishitani (2009). Se
ond, protability and pressure 45 Table 3.5: Industry 
omposition of air emitting fa
ilities with an average of main variables Name of industry Num of fa
i. ISO 14001 E A Fa
i. w/ E W = 0 Workers Crude oil and natural gas produ
tion 21 47.6% 2.16 21 22.43 Food 13 46.2% 1.99 11 248.77 Textile mill produ
ts Mfg. 44 25.0% 5.05 29 219.09 Apparel and other nished produ
ts made from fabri
s and similar materials Mfg. 11 0.0% 0.129 10 104.55 Lumber and wood produ
ts, ex
ept furniture Mfg. 38 78.9% 0.787 38 117.95 Furniture and xtures Mfg. 30 60.0% 0.453 27 157.50 Pulp, paper and paper produ
ts Mfg. 93 40.9% 2.48 68 238.76 Printing and allied industries 143 17.5% 0.829 142 187.20 Chemi
al and allied produ
ts Mfg. 856 39.0% 5.47 657 173.39 Petroleum and Coal produ
ts Mfg. 64 54.7% 0.706 56 191.61 Plasti
 Produ
ts Mfg. 300 38.0% 1.15 279 173.63 Rubber Produ
ts Mfg. 108 44.4% 2.62 99 404.13 Cerami
, stone and 
lay produ
ts Mfg. 118 36.4% 3.80 84 294.17 Iron and steel Mfg. 86 47.7% 9.78 58 623.98 Non-ferrous metals and produ
ts Mfg. 95 36.8% 7.62 67 273.21 Fabri
ated metal produ
ts Mfg. 423 25.1% 0.930 377 165.61 General ma
hinery Mfg. 239 51.0% 0.755 232 562.54 Ele
tri
al ma
hinery, equipment and supplies Mfg. 314 67.8% 0.559 239 751.16 Transportation equipment Mfg. 395 54.2% 1.56 328 936.69 Pre
ision instruments and ma
hinery Mfg. 61 39.3% 1.03 51 383.30 Mis
ellaneous manufa
turing industries 127 39.4% 0.525 121 179.11 from residents of the lo
al 
ommunity are not determinants of ISO 14001 adoption by fa
ilities that emit into water. Third, muni
ipalities' support for ISO 14001 adoption does not stimulate ISO 14001 adoption; fa
ilities lo
ated in muni
ipalities that provide support for ISO 14001 adoption are not more likely to adopt the standards than those in muni
ipalities that do not provide support. Finally, dirty (high emission intensity) fa
ilities are less likely to adopt ISO 14001 to improve their environmental performan
e, but 
lean fa
ilities are more likely to adopt the standards as a sign that they are (a
tually) 
lean. Spillovers (n) are 
ompletely different between models. The spillover effe
ts between large num- bers of fa
ilities are negative (WM II and III models), whereas the spillover effe
ts between fa
ili- ties with similar 
hara
teristi
s are positive (WM IV and V models). Ex
ept for the WM II model, where the spillover effe
t is signi
ant at the 5% level, these spillover effe
ts are signi
ant at the 1% level. Negative spillover effe
ts (WM II and III) might ree
t that the different-sized fa- 46 Table 3.6: Estimation results of ISO 14001 adoption by water emitting fa
ilities Dependent variable: ISO 14001 adoption WM I WM II WM III WM IV (m=10) WM V (m=10) Workers/10 2 0.142  0.140  0.136  0.094  0.109  (0.011) (0.011) (0.012) (0.015) (0.014) Air emission intensity/10 6 0.025 0.03 0.043 0.075 0.029 (0.248) (0.246) (0.251) (0.271) (0.254) Water emission intensity/10 6 -0.200  -0.178  -0.224  -0.137  -0.147  (0.112) (0.090) (0.123) (0.076) (0.084) Prot/revenue/10 3 1.04 1.00 1.10  0.912 0.923 (0.603) (0.666) (0.624) (0.652) (0.665) Population density/10 3 -4.78 -4.78 -4.64 -3.81 -4.36 (3.36) (3.27) (3.20) (3.57) (3.56) In
ome per 
apita/10 3 0.308 0.331 0.367 0.223 0.242 (0.359) (0.372) (0.360) (0.362) (0.376) Support 0.007 -0.028 -0.028 -0.032 -0.016 (0.132) (0.140) (0.134) (0.140) (0.144) Grievan
es against pollution -0.664 -0.596 -0.603 -0.612 -0.481 (0.456) (0.454) (0.453) (0.468) (0.427) Year dummy (2001) -0.275  -0.436  -0.439  -0.200  -0.225  (0.083) (0.132) (0.122) (0.073) (0.078) Constant -13.3  -8.53 -21.0  -29.6  -7.442 (7.52) (5.38) (9.14) (10.5) (4.78) n -0.639  -0.669  0.334  0.166  (0.282) (0.188) (0.067) (0.054) Log-likelihood -1474.06 -1464.54 -2245.84 -1121.92 -1099.74 BIC -1711.33 -1705.4 -2486.7 -1362.78 -1340.60 AIC -1540.06 -1531.54 -2312.84 -1188.92 -1166.74 LR test statisti
 19.0  -1543.6 704.3  748.6  Observation 1326 1326 1326 1326 1326 All of models in this table are SAR probit models. Standard errors are shown in parentheses.  ,  , and  imply that the 
oef
ient is signi
antly different from zero at the 10%, 5% and 1% levels, respe
tively. All models are estimated with industry and prefe
ture dummies. 
ilities have different in
entives to adopt the ISO 14001 standard. Large fa
ilities have positive attitude toward ISO 14001 adoption, while small fa
ilities have negative attitude. The "n"s of WM II and III are the average effe
ts of the industry-wide and the 
ountry-wide attitude (aggregate atti- tude) toward ISO 14001 adoption on the individual fa
ilities' attitudes, respe
tively. As mentioned previously, the ISO 14001 adoption rate of our sample is 58.5%, and the adoption rates of most industries are 40-60%. Therefore, approximately half of the fa
ilities have attitudes toward this adoption that differ from the aggregate attitude. Negative spillover effe
ts are likely to ree
t this 47 Table 3.7: Estimation results of water emissions redu
tions Dependent variable: Water emissions redu
tions (E W t+1 =E W t ) WM I WM II WM III WM IV (m=3) WM V (m=3) Workers -0.367 -0.366 -0.375 -0.367 -0.367 (1.73) (1.73) (1.73) (1.73) (1.73) Prot/Revenue -0.351 -0.351 -0.337 -0.352 -0.351 (4.52) (4.51) (4.52) (4.51) (4.52) ISO 172 172 197 176 172 (849) (848) (849) (847) (849) Population density -1167 -1166 -1225 -1171 -1167 (1746) (1743) (1745) (1742) (1746) In
ome per 
apita 2.20 2.20 1.77 2.22 2.20 (12.1) (12.1) (12.1) (12.1) (12.1) Grievan
es against pollution -1.90 -1.90 -2.40 -1.92 -1.90 (10.2) (10.2) (10.2) (10.1) (10.2) Constant -135 -347 -185 -134 -135 (488) (488) (488) (487) (488) r -0.990 -0.153 0.006 0.0003 (0.000) (0.000) (0.000) (0.0000) Log-likelihood -6647.041 -6646.352 -6646.850 -6647.038 -6647.041 BIC -6669.780 -6672.339 -6672.837 -6673.025 -6673.028 AIC -6654.041 -6654.352 -6654.850 -6655.038 -6655.041 LR test statisti
 1.38 0.382 0.006 0.000 Observation 663 663 663 663 663 All of models in this table are xed effe
ts SAR models.  ,  , and  imply that the 
oef
ient is signi
antly different from zero at the 10%, 5% and 1% levels, respe
tively. differen
e in attitude toward ISO 14001 adoption. Whi
h effe
t more effe
tively explains the ISO 14001 adoption of water emitting fa
ilities? To 
ompare the spe
i
ations, we 
al
ulate BIC and AIC, a

ording to Hepple (2004) 6 . Be
ause the BIC and AIC differen
e between V and others are greater than 10, we have very strong eviden
e that the WM V model, or the intra-industry spillover between fa
ilities of similar sized rms (in terms of similar revenue) has better explanatory power for ISO 14001 adoption by fa
ilities that emit into water. This result is quite intuitive. Be
ause the adoption of ISO 14001 demands a large initial 
ost, the de
ision to do so is less likely to be made by the fa
ilities than by the rms to whi
h they belong. 6 The BIC and AIC used here are BIC=log-likelihood(the number of 
oef
ient)ln(Observation)/2, and AIC=log-likelihood(the number of 
oef
ient). 48 Table 3.8: Estimation results of negative binomial regressions for grievan
es against pollution (wa- ter emitting fa
ilities) Dependent variable: Grievan
es against pollution 
onstant mean 
onstant mean Per
entage 
hange in water emissions (E W t+1 =E W t ) 4.95E-6 3.10E-6 (3.63E-6) (9.00E-6) Weighted sum of emissions into water 1.40E-10  8.47E-11 (5.12E-11) (1.17E-10) Constant 4.24  4.24  4.24  4.24  (0.04) (0.04) (0.04) (0.04) Log-likelihood -6650.51 -6649.96 -6650.25 -6648.34 LR test of a = 0 or d = 0 1.4E+5  1.4E+5  1.4E+5  1.4E+5  Observation 1326 1326 1326 1326 "
onstant” implies that we employ a negative binomial model with 
onstant-dispersion for estimation, and "mean" implies that we employ a negative binomial model with mean-dispersion. Standard errors are shown in parentheses.  ,  , and  imply that the 
oef
ient is signi
antly different from zero at the 10%, 5% and 1% levels, respe
tively. all models are estimated with industry and prefe
ture dummies. Similarly sized rms in the same industry (when revenue is used as a proxy for rm size) are likely to be 
ompetitors, and a model of intra-industry spillover effe
ts between 
ompetitors 
an explain the data most effe
tively. Therefore, imitation pressure be
ause of 
ompetition has likely 
ontributed to the diffusion of ISO 14001 in Japan. Table 3.7 shows the determinants of water emissions 
hanges. We use the values of ISO 14001 predi
ted by ISO 14001 adoption estimation with weight matrix V. All explanatory variables are in signi
ant at any level, and therefore, ISO 14001 adoption does not redu
e water emission intensity. In addition, BIC and AIC provide little eviden
e that models with 
orrelations of water emission redu
tions have better explanatory power than models without the 
orrelation, although ris signi
ant at 1% level. Fa
ilities with greater water emissions are likely to be the target of more grievan
es. Therefore, there may be an endogeneity problem in emissions redu
tion equations in that environmental per- forman
e affe
ts grievan
es, and this endogeneity 
ould bias the estimated 
oef
ients. To assess whether an endogeneity problem exists, we regress the grievan
es variable separately on the per- 
entage 
hange in water emissions and on the weighted sum of water emissions. We employ a negative binomial model for the regressions. The results of these regressions are presented in Table 49 3.8. These results indi
ate that the grievan
es might be affe
ted by the weighted sum of emissions but are not affe
ted by per
entage 
hange in water emissions, whi
h we employ as an indi
ator of environmental performan
e. Thus, there is no endogeneity due to the 
ausal effe
t of environ- mental performan
e on grievan
es, and the estimation results of water emission redu
tions are not affe
ted by su
h endogeneity. 3.5.2 Performan
e and ISO 14001 adoption of fa
ilities that emit into air Table 3.9 shows the estimates of ISO 14001 adoption by fa
ilities that emit into air. The number of workers in a fa
ility has a signi
antly positive effe
t on ISO 14001 adoption at the 1% level under all spe
i
ations, whereas grievan
es against pollution have a negative effe
t at the 10% level. However, emission intensity, support for ISO 14001 adoption, protability, and other 
hara
teristi
s of the muni
ipality in whi
h a fa
ility is lo
ated do not have signi
ant effe
ts. This result is the same as the estimation results of fa
ilities that emit into water. Unlike fa
ilities that emit into water, a fa
ility lo
ated in a muni
ipality with fewer grievan
es against pollution is more likely to adopt and (air) emission intensity does not. Muni
ipalities with fewer grievan
es against pollution 
an be interpreted as 
leaner ones. Fa
ilities in su
h muni
ipalities are under more pressure from a muni
ipal government or from residents to adopt environmental pra
ti
es, so the estimation result for grievan
es against pollution may imply that pressure from a muni
ipal government or from residents has an impa
t on ISO 14001 adoption de
isions. Similarly to the fa
ilities that emit into water, the spillover effe
ts between large numbers of fa- 
ilities are negative (WM II and III models), whereas the spillover effe
ts between fa
ilities with similar 
hara
teristi
s are positive (WM IV and V models). With the ex
eption of the WM II model, where the spillover is signi
ant at the 10% level, these spillover effe
ts are signi
ant at the 1% level. Similarly to the fa
ilities that emit into water, the negative spillover effe
ts (WM II and II I) for the fa
ilities that emit into air might ree
t that the different-sized fa
ilities have different in
entives to adopt the ISO 14001 standard. This is be
ause roughly half of the fa
ilities have dif- ferent attitude toward the adoption than the aggregate attitude, as is indi
ated from the fa
t that ISO 14001 adoption rates of our sample and of most industries are 41.4% and 40-60%, respe
tively. 50 Table 3.9: Estimation results of ISO 14001 adoption by fa
ilities emitting into air Dependent variable: ISO 14001 adoption WM I WM II WM III WM IV (m=10) WM V (m=10) Workers/10 2 0.145  0.143  0.143  0.097  0.085  (0.005) (0.006) (0.005) (0.016) (0.006) Air emission intensity/10 6 0.148 0.169 0.159 0.163 0.165 (0.129) (0.127) (0.125) (0.130) (0.125) Water emission intensity/10 6 -0.121 -0.121 -0.101 -0.139 -0.118 (0.109) (0.196) (0.103) (0.107) (0.102) Prot/revenue/10 3 0.159 0.079 0.064 -0.05 0.061 (0.227) (0.230) (0.232) (0.220) (0.233) Population density/10 3 -1.72 -1.69 -1.74 -1.95 -1.20 (1.15) (1.13) (1.07) (1.24) (1.17) In
ome per 
apita/10 3 -0.11 -0.112 -0.121 -0.216 -0.129 (0.112) (0.115) (0.112) (0.123) (0.121) Support 0.069 0.071 0.074 -0.054 0.059 (0.049) (0.049) (0.047) (0.046) (0.048) Grievan
es against pollution -0.293 -0.305  -0.293  -0.235 -0.355  (0.169) (0.166) (0.161) (0.182) (0.171) Year dummy (2001) -0.209  -0.316  -0.375  -0.130  -0.131  (0.319) (0.072) (0.050) (0.035) (0.028) Constant -15.6  5.12  -11.9  -54.5  -3.02  (4.67) (2.17) (5.27) (21.4) (1.26) n -0.544  -0.847  0.430  0.417  (0.305) (0.097) (0.090) (0.025) Log-likelihood -6018.23 -6016.8 -6027.13 -6352.29 -6337.96 BIC -6359.96 -6358.53 -6368.86 -6694.01 -6679.68 AIC -6095.23 -6093.80 -6104.13 -6429.29 -6414.96 LR test statisti
 2.86  -17.8 -668.1 -638.5 Observation 7158 7158 7158 7158 7158 All of models in this table are SAR probit models. Standard errors are shown in parentheses.  ,  , and  imply that the 
oef
ient is signi
antly different from zero at the 10%, 5% and 1% levels, respe
tively. All spe
i
ations are estimated with industry and prefe
ture dummies. Among the models with a spatial 
orrelation, WM II has the greatest BIC and AIC. Compared with the WM I model, the WM II model has an AIC that is greater by approximately 1 but a BIC that is smaller by approximately 3. Therefore, we do not have positive eviden
e that the spe
i
ation with the spillover effe
t performs better. The determinants of air emission 
hanges are presented in Table 3.10. We use values of ISO 14001 predi
ted by ISO 14001 adoption estimation with Weight Matrix I (no spatial 
orrelation model). All of the explanatory variables are insigni
ant at all levels, and therefore, ISO 14001 adoption 51 Table 3.10: Estimation results of air emissions redu
tions (SAR) Dependent variable: Air emissions redu
tions (E A t+1 =E A t ) WM I WM II WM III WM IV (m=5) WM V (m=10) Workers 0.0248 0.0248 0.0276 0.0249 0.0256 (0.0528) (0.0528) (0.0528) (0.0515) (0.0526) Prot/Revenue 0.0163 0.0163 0.068 0.0069 0.0189 (0.100) (0.100) (0.100) (0.098) (0.100) ISO -13.1 -13.1 -14.7 -13.8 -13.3 (20.6) (20.6) (20.6) (20.1) (20.5) Population density 6.41 6.41 5.81 5.21 5.97 (26.7) (26.7) (26.7) (26.1) (26.6) In
ome per 
apita -0.178 -0.178 -0.183 -0.169 -0.176 (0.233) (0.233) (0.232) (0.227) (0.232) Grievan
es against pollution 0.144 0.144 0.13 0.131 0.142 (0.202) (0.202) (0.201) (0.197) (0.201) Constant 6.72 16.4 3.69 6.06 6.36 (10.8) (10.8) (10.7) (10.5) (10.7) r -0.990  0.318  0.110  0.0390  (0.000) (0.000) (0.000) (0.000) Log-likelihood -25426.65 -25426.96 -25423.02 -25419.32 -25426.18 BIC -25455.29 -25459.69 -25455.75 -25452.06 -25458.91 AIC -25433.65 -25434.96 -25431.02 -25427.32 -25434.18 LR test statisti
 -0.620 7.26  14.7  0.940 Observation 3579 3579 3579 3579 3579 All of models in this table are xed effe
ts SAR models.  ,  , and  imply that the 
oef
ient is signi
antly different from zero at the 10%, 5% and 1% levels, respe
tively. does not redu
e air emission intensity as it does water emission intensity. However, r is signi
ant at the 1% level for all models with spillover effe
ts, and BIC and AIC provide positive eviden
e that the Weight Matrix IV model has better explanatory power than one without the 
orrelation. This result implies that the per
entage 
hange in air emissions of similarly sized fa
ilities in the same industry are 
orrelated; i.e., a fa
ility redu
es its air emissions if similarly sized fa
ilities do so. This 
orrelation may ree
t te
hnology diffusion; similarly sized fa
ilities in the same industry are likely to have similar te
hnology. In 
ontrast with air emitting fa
ilities, the low explanatory power of spillover models among similarly sized rms and fa
ilities for water emission redu
tions may ree
t the fa
t that water emitting fa
ilities have different te
hnology even though they are similar in size. Is r for the air emission redu
tion equation (parti
ularly, in WM IV model) signi
ant be
ause 52 Table 3.11: Estimation results of air emissions redu
tions (SAR and SAC) Dependent variable: Air emissions redu
tions (E A t+1 =E A t ) SAC SAR SAC FS WM I FS WM II FS WM II m=5 m=5 m=5 Workers 0.0249 0.0248 0.0248 (0.0515) (0.0515) (0.0515) Prot/Revenue -0.00693 -0.00686 -0.00679 (0.0980) (0.0980) (0.0980) ISO -13.8 -13.0 -13.0 (20.1) (20.0) (20.0) Population density 5.21 5.15 5.15 (26.1) (26.1) (26.1) In
ome per 
apita -0.169 -0.169 -0.169 (0.227) (0.227) (0.227) Grievan
es against pollution 0.131 0.131 0.131 (0.197) (0.197) (0.197) Constant 6.06E 5.99 6.00 (8.69) (10.5) (8.71) r 0.110  0.109  0.109  (0.027) (0.000) (0.027) t -0.999 -0.999 (1.41) (1.41) Log-likelihood -25418.62 -25419.34 -25418.67 Observation 3579 3579 3579  ,  , and  imply that the 
oef
ient is signi
antly different from zero at the 10%, 5% and 1% levels, respe
tively. FS WM I and II imply we use values of ISO 14001 predi
ted by the rst stage estimation with weight matrix I and II, respe
tively. emission redu
tions are a
tually 
orrelated or only be
ause the errors are 
orrelated? To examine this question, we employ Spatial Auto-Correlation(SAC) models to examine whether r (
orrelation in air emissions redu
tions between fa
ilities) for the air emission redu
tion equation is signi
ant be
ause emission redu
tions are a
tually 
orrelated or only be
ause errors are 
orrelated. A typi
al SAC model is as follows: y= rW 1 y+Xb +u u= tW 2 u+ e e  N(0;s I n ) (3.11) W 1 andW 2 
an be the same or different. If r = 0, this model is the same as a Spatial Error Model (SEM). First, we estimate SEMs to sele
t W 2 from the ve weight matri
es used for the SAR 53 Table 3.12: Estimation results of negative binomial regressions for grievan
es against pollution (air emitting fa
ilities) Dependent variable: Grievan
es against pollution 
onstant mean 
onstant mean Per
entage 
hange in water emissions (E A t+1 =E A t ) -3.59E-6 -1.62E-5 (4.67E-5) (6.24E-5) Weighted sum of emissions into water 1.81E-9  1.80E-12 (6.67E-10) (1.18E-09) Constant 4.36  4.36  4.35  4.36  (0.02) (0.02) (0.02) (0.02) Log-likelihood -36486.11 -36486.06 -36483.01 -36486.11 LR test of a = 0 or d = 0 8.8E+5  8.8E+5  8.8E+5  8.8E+5  Observation 7158 7158 7158 7158 "
onstant” implies that we employ a negative binomial model with 
onstant-dispersion for estimation, and "mean" implies that we employ a negative binomial model with mean-dispersion. Standard errors are shown in parentheses.  ,  , and  imply that the 
oef
ient is signi
antly different from zero at the 10%, 5% and 1% levels, respe
tively. all models are estimated with industry and prefe
ture dummies. models. Be
ause the likelihood of the WM I model is the greatest, we sele
t WM I forW 2 . Then, we estimate the SAC models by settingW 2 to weight matrix II. In addition, we estimate equations for the per
entage 
hange in air emissions with values of the ISO 14001 adoption predi
ted by the model with WM II. Table 3.11 shows estimates of the SAC model with the value of ISO 14001 predi
ted by ISO 14001 adoption estimation with weight matrix I in the rst 
olumn, and the SAR and SAC model with the value of ISO 14001 predi
ted by ISO 14001 adoption estimation with weight matrix II in the se
ond and third 
olumns, respe
tively. We use weight matrix IV for all these models. A

ording to Table 3.11 and the last 
olumn of Table 3.10, r is signi
ant and its value 
hanges little, even with error 
orrelation. This result indi
ates that per
entage 
hanges in emission redu
tions are a
tually 
orrelated. As with the water-emitting fa
ilities, we assess whether an endogeneity problem arises for the air-emitting fa
ilities be
ause of the 
ausal effe
t of environmental performan
e on grievan
es by regressing the grievan
es variable separately on the per
entage 
hange in air emissions and on the weighted sum of air emissions. Table 3.12 presents the results of these regressions. These results indi
ate that the estimation results of air emissions redu
tions are not affe
ted by su
h endogeneity. 54 3.6 Con
lusion In the OECD surveys, many fa
ilities evaluate the pra
ti
es “other fa
ilities like ours is adopt- ing ” as “very important” or “important”as a motivation for adopting an EMS and environmental pra
ti
es. This 
hapter examines su
h externalities in ISO 14001 adoption and environmental per- forman
e, and it also examines the determinants of ISO 14001 adoption and environmental per- forman
e by employing spatial e
onometri
 estimation methods. We fo
us on Japanese fa
ilities be
ause Japan has ranks rst worldwide with the most fa
ilities that have adopted ISO 14001. We nd the following results for ISO 14001 adoption. First, the number of workers in a fa
ility has signi
antly positive effe
ts on ISO 14001 adoption, as was found in previous studies on ISO 14001. Se
ond, fa
ilities that emit into water with lower water emission intensity are more likely to adopt ISO 14001. Third, the ISO 14001 adoption by fa
ilities that emit into water 
an be more effe
tively explained by the intra-industry spillover effe
ts between fa
ilities of rms with similar revenue than other spillover effe
ts, whereas the intra-industry spillover effe
ts 
annot more effe
- tively explain the adoption of fa
ilities that emit into air. Be
ause we employ revenue as a proxy of rm size, similar-sized rms make similar de
isions on ISO 14001 adoption. In addition, this result may indi
ate that the ISO 14001 adoption de
isions of 
ompetitors inuen
e ea
h other as similar-sized rms in the same industry are likely to be 
ompetitors, The results of this 
hapter suggest that ISO 14001 adoption does not have a signi
ant impa
t on environmental performan
e or on other fa
tors. However, the per
entage 
hange in air emissions, whi
h was used as a proxy for environmental performan
e, is 
orrelated between similar sized fa- 
ilities in the same industry. This 
orrelation might ree
t te
hnology diffusion be
ause similar sized fa
ilities in the same industry are likely to use similar te
hnology. Our estimation results sug- gest the possibility that similar-sized rms or fa
ilities implement similar voluntary environmental a
tions and that there are unlikely spillover effe
ts of voluntary environmental a
tions between fa
ilities and workers with various 
hara
teristi
s. Our nding son spillover effe
ts suggest that governments might be able to pursue voluntary programs of industry-spe
i
 or narrower targets in 
ooperation with industry asso
iations. We fo
us on intra-industry spillover effe
ts of voluntary environmental a
tions in Japan. However, 55 geographi
al spillover effe
ts of voluntary environmental a
tions might also exist and may be 
or- related with other spillover effe
ts. Therefore, it would be interesting to examine these two types of spillover effe
ts simultaneously, although an estimation in
luding two spillover effe
ts would be mu
h more 
ompli
ated and 
omputation intensive. Examining multiple simultaneous spillover effe
ts remains a topi
 for future resear
h. 56 Chapter 4 Taxes versus Quotas in Lobbying by a Polluting Industry with Private Information on Abatement Costs 4.1 Introdu
tion Two of the main 
auses of inef
ien
y of environmental regulations are informational issues about polluters' abatement 
ost and politi
al pressure by polluting industries when the regulator 
an spe
- ify how mu
h parti
ular industries are permitted to pollute. Many resear
hers have studied these two 
auses and their impa
ts on environmental regulation. For informational issues, Weitzman (1974) rst 
ompared taxes and quotas in the presen
e of un- 
ertainty about marginal 
osts and benets in a general 
ontext. Sin
e Weitzman, many resear
hers have 
ompared these two instruments for environmental prote
tion under un
ertainty about abate- ment 
ost in different settings 1 . Moledina et al. (2003) and Costello and Karp (2005) fo
us on asymmetri
 information about abatement 
osts and 
ompare taxes and permits in a dynami
 set- ting. While the rms take advantage of asymmetri
 information and behave strategi
ally in the model of Moledina et al. (2003) the rm in Costello and Karp (2005). is non-strategi
. Other pa- pers have modeled a pollution 
ontrol problem under asymmetri
 information as a prin
ipal-agent or me
hanism design problem. Dasgupta et al. (1980) applied a Groves me
hanism to a multiple- 1 For example, several papers 
ompare two instruments for sto
k pollutant 
ontrol (Hoel and Karp (2001), Hoel and Karp (2002), Karp and Zhang (2002), and Newell and Pizer (2003)). Quirion (2004) analyzes a 
ase with pre-existing distortionary taxes, and Kaplow and Shavell (2002)analyze a 
ase in whi
h a regulator 
an impose a nonlinear tax 57 polluter regulation problem, while Lewis (1996) and other studies employed the prin
ipal-agent framework. In these frameworks, a government implements a 
ombination of pollution quotas and taxes (or subsidies). Resear
h on the politi
al e
onomy of environmental poli
ies began re
ently relative to resear
h on informational issues. By applying to environmental poli
ies a lobbying model developed byGross- man and Helpman (1994), Fredriksson (1997) and Aidt (1998) independently examine how pollu- tion taxes are affe
ted by lobbying groups. Damania (2001) shows if rms 
an politi
al inuen
e the goverrnment, then they underinvest in 
lean te
hnologies to make lobby against environmen- tal poli
y effe
tive. Con
oni (2003) analyzes how a green lobby affe
ts trade and environmental poli
ies in two large e
onomies with transboundary pollution. Barbier et al. (2005) theoreti
ally and empiri
ally examine how dynami
 resour
e 
onservation poli
ies are affe
ted by industry lob- bies. Yu (2005) analyzes a model where lobbying groups affe
t publi
 environmental awareness by advertising before they lobby the government. These papers assume that a government knows the pollution abatement 
ost of industry. The above studies take into a

ount only one of the two 
auses of inef
ient regulation. However, governments usually fa
e both informational and politi
al e
onomy issues. This 
hapter 
ompares taxes and quotas when a government fa
es both issues or is affe
ted by an industry lobby with pri- vate information about its abatement 
ost. Espe
ially, we examine the effe
ts of private information about abatement 
ost on lobbying a
tivity, environmental regulation and so
ial welfare and show how different these effe
ts are under tax and quota systems. To do so, we in
orporate the prin
ipal- agent relationship with an informed prin
ipal and 
ommon-value private information examined by Maskin and Tirole (1992) 2 into a lobby model of the Grossman and Helpman type. Our results differ from the literature. First, our result on so
ial welfare is 
ompletely different from the results in the literature on taxes versus quotas under imperfe
t information. In the literature on “tax versus quota”, a tax is basi
ally 
onsidered better than a quota if the marginal abatement 
ost 2 In general, we 
an 
lassify the prin
ipal's private information into one of two forms based on whether the prin
i- pal's private information is an argument of the agent's obje
tive fun
tion. If it is an argument, the private information is 
lassied as 
ommon value. Otherwise, the private information is 
lassied as private value. Be
ause a government of this 
hapter's model, an agent, takes into a

ount abatement 
ost, like most models of lobbying on environmental pol- i
y, private information is an argument of the agent's obje
tive fun
tion. For this reason, we employ a 
ommon-value model. 58 (MAC) 
urve's slope is steeper than the marginal damage (MD) 
urve's slope. In our model, how- ever, quotas are so
ially preferred (1) when the government weighs politi
al 
ontributions mu
h more strongly than it does so
ial welfare or (2) when the MAC 
urve's slope is steeper than the MD 
urve's slope. The industry generally has a stronger in
entive to, and a
tually does, relax reg- ulations under taxes than under quotas due to tax payments. This nding is parti
ularly the 
ase when the government is easily affe
ted by politi
al 
ontributions be
ause paying politi
al 
ontri- butions for low tax rates is 
heaper than paying tax under su
h a 
ase. The industry's benet from lobbying is larger under taxes than it is under quotas if the slope of the MAC 
urve is steep (relative to the MD 
urve). Consequently, quotas are better than taxes when the MAC 
urve's slope is steep. Be
ause a steep slope implies that a pri
e 
hanges more than the quantity does, a redu
tion in the abatement target lowers the tax rate and tax payments more than the (a
tual) abatement does when the MAC 
urve's slope is steep. Therefore, in su
h a 
ase, the industry's 
ost savings due to the redu
tion of the abatement target by one unit is greater under taxes than it is under quotas. Therefore, the government is under high politi
al pressure from the industry, and it sets a low tax rate. The impa
t of private information on so
ial welfare under our model might also be different from that under standard regulation models. It is well known that if a polluting industry 
annot polit- i
ally inuen
e environmental regulations (standard regulation models), a government has to pay an information rent to some type(s) of industry that has an in
entive to pretend that it is of another type(s). However, if the industry 
an do so, it may have to reveal its own type (on abatement 
ost), with some 
ost. For example, when the tax is implemented, a low-
ost industry (i.e., low abatement 
ost) will fa
e a lower tax rate and lower politi
al 
ontribution than a high-
ost industry (be
ause the low-
ost industry will abate enough even if its tax rate is low relative to that of the high-
ost industry). Therefore, the high-
ost industry has an in
entive to pretend to be the low-
ost industry, while the low 
ost industry evades higher tax rate and politi
al 
ontribution by revealing its type. Relative to the high-
ost industry, the low-
ost industry 
an endure a high tax rate be
ause it does not emit mu
h, but the high-
ost industry has higher willingness to pay for a low tax rate 
ompared to the low-
ost industry. The low-
ost industry has to a

ept a higher tax rate in order to reveal its type. The higher tax rate 
onstitutes a 
ost to the industry, but it is a benet for so
iety (provided 59 that the tax rate is not too high). Thus, private information in a 
ase with politi
al inuen
e of the polluting industry has different impa
ts 
ompared to private information in a 
ase without politi
al inuen
e. A
tually, what happens under a tax depends on whether the slope of MAC 
urve is steeper than the respe
tive slopes of the (a
tual) MD 
urve and the weighted MD 
urve (MD 
urve multiplied by government's weight on so
ial welfare) 3 . We show that if the MAC 
urve's slope is steeper than the weighted MD 
urve's slope, the private information might improve so
ial welfare by the me
hanism dis
ussed in the last paragraph. However, if the MAC 
urve's slope is steeper than the MD 
urve's slope, the low-
ost industry might not be able to reveal its type. Be
ause the differen
e in tax rates between the low- and high-
ost industries is great in su
h a 
ase (as pri
e 
hanges more than quantity does), the high-
ost industry has a strong in
entive to pretend to be the low-
ost industry in order to save abatement 
ost and redu
e its tax payment. Therefore, it is very hard or impossible for the low-
ost industry to reveal its type, and a separating equilibrium where the government 
an differentiate the industry's type might not exist. In 
ontrast to the 
ase of taxes, private information does not improve so
ial welfare when quotas are implemented. This is be
ause that the high-
ost industry might have to make the government set a higher quota than it would under 
omplete information to reveal its 
ost information. However, a separating equilibrium always exists, unlike in the 
ase of taxes. The quota for the high-
ost industry under a 
omplete information 
ase is higher than that for the low-
ost industry, and the high-
ost industry is more in favor of a high quota 
ompared to the low-
ost industry. Therefore, by making the government set a higher quota than would be set under the 
omplete information 
ase, the high-
ost industry 
an reveal its type to the government. Thus, private information redu
es the 
omparative disadvantage of taxes 
ompared to quotas when the government 
ares little about so
ial welfare. Although environmental groups also lobby the government, it appears that analyzing only the in- dustry lobby with private information is not unreasonable. During the 1990s, 
ontributions to the U.S. Congress from ele
tri
 utilities and the oil and gas industry alone were roughly 10 times larger 3 “The MAC 
urve slope is steeper than the MD 
urve slope or the weighted MD 
urve slope” means that the abatement 
ost 
hanges more than the damage or weighted damage 
aused by a one-unit 
hange in emissions. 60 than the 
ontributions from environmental organizations a

ording to data from the Center for Re- sponsive Politi
s. In addition, there were no politi
al 
ontributions from pro-environmental poli
y groups in 10 states in the US, and the total 
ontribution from the energy industry was at least 10 times larger than that from pro-environmental poli
y groups in most states during 2003-2006 4 . In this 
hapter, the industry lobby group indire
tly transmits its private information to the gov- ernment. Therefore, the paper might fall within an informational lobbying literature that analyzes the transmission of information from lobby group(s) to a poli
y-maker. Some papers in the liter- ature analyze 
ases with two (or three) poli
y alternatives (e.g., Austen-Smith and Wright (1994), Lohmann (1995), Bennedsen and Feldmann (2006), Dahm and Porteiro (2008)). Other papers ex- amine the 
ase when the government 
hooses a poli
y from a 
ontinuous set of alternatives, as in this 
hapter (e.g., Austen-Smith (1995), Austen-Smith and Banks (2002)). A 
ontinuous set of alternatives (regulation levels) is preferable when analyzing how the regulation level is affe
ted by lobbying. However, in papers with 
ontinuous sets of alternatives, the poli
y must be very abstra
t to enable the analysis of strategi
 transmission of information. In addition, a

ording to the form of the utility fun
tion, every type has a different preferen
e for poli
y in these papers, and every type of industry wants a lower emission tax or a higher emission quota. Thus, the settings of these papers are not suitable to analyze lobbying for environmental poli
ies by polluting industries possessing private information. Finally, we have to mention Boyer and Laffont (1999) who, like us, examine environmental poli
ies under a 
ase with politi
al e
onomy and informational issues. They 
ompare the in
entive me
h- anism (abatement level and transfer) with a single abatement level based on expe
ted abatement 
ost when a monopoly has private information on pollution abatement 
ost and politi
al majorities representing different stakes 
hooses the instrument. They show that the in
entive me
hanism may not be desirable if there are more stakeholders in the monopoly in a so
iety or if informational rents to the stakeholders are large. However, they fo
us only on quantity regulations (dire
t quan- tity 
ontrol regulations), whereas this 
hapter 
ompares quantity and pri
e instruments (dire
t and indire
t 
ontrol regulations). 4 There are two ex
eptions. In Alabama, 
ontributions from the energy industry amounted to $811,300, whereas 
ontributions from pro-environmental poli
y groups amounted to $421,409. In Oregon, $993,038was 
ontributed from the energy industry and $260,278 from pro-environmental poli
y groups. See Moore (2007) for further detail. 61 This 
hapter is organized as follows. Se
tion 2 des
ribes the model environment. Se
tion 3 argues the 
omplete information 
ase, while se
tion 4 argues the in
omplete information 
ase. Se
tion 5 
ompares the 
onsequen
es of a tax with those of a quota using numeri
al methods. We 
on
lude in se
tion 6. 4.2 Setup There are two types of industries, one with low abatement 
ost and one with high abatement 
ost, denoted by i= L;H. Let the abatement 
ost of type i be C i (e i ) =C(e i ; ē i ) = 8 > < > : 1 2 
(ē i  e i ) 2 i f ē i  e i 0 i f ē i < e i (4.1) for i = L;H where e i is emission by type i and ē i 
an be private information held by the industry. 
; ē i , and ē i e i 
an be interpreted as the slope of the marginal abatement 
ost (MAC) 
urve, natural emission of type i and pollution abatement of type i, respe
tively. The industry of type i makes a politi
al 
ontribution, W i , to make an environmental regulation (quota or tax) more lenient. The total 
ost of the industry of type i is C i (e i (q))+W i (q) under a quota and isC i (e i (t))+ te i (t)+W i (t) under a tax, where q is the quota and t is the tax rate. Damage by pollution is denoted by D(e) = 1 2 e 2 . Therefore, a benevolent poli
ymaker set an en- vironmental regulation to minimize the so
ial 
ost, total pollution abatement 
ost and damage by pollution, SC i (e i ) = D(e i )+C i (e i ). However, poli
y-making will be inuen
ed by lobbying by the industry. We follow Grossman and Helpman and assume that the obje
tive fun
tion of the govern- ment isW i (t)aSC i (e i (t)) when it implements a tax andW i (q)aSC i (e i (q)) when implementing a quota, where a > 0 is the exogenously given weight of so
ial welfare relative to the 
ontribution. Politi
al pro
ess There are two stages in the politi
al pro
ess between the government and industry. In the rst stage, the industry offers the government a 
ontingent 
ontribution s
hedule W i (t) in the tax 
ase 62 and W i (q) in the quota 
ase. W i (t) and W i (q) are 
ontinuously differentiable fun
tions mapping from T to R + and from Q to R + where T = [0;
ē H ℄ and Q = [0; ē H ℄ are one-dimensional tax and quota 
hoi
e set, respe
tively. In the se
ond stage, the government sele
ts a regulation level (tax rate or quota) and re
eives from the industry the 
ontribution asso
iated with the sele
ted regulation level. 4.3 Complete information 
ase We 
hara
terize the equilibria under the 
omplete information 
ases as follows. If the government implements a tax, (fW C i ; t C i g i=L;H ) is the Subgame Perfe
t Nash Equilibrium (SPNE) if (i) t C i maximizes W C i (t i )aSC i (t i ) (ii) t C i minimizes C i (e i (t i ))+ t C i e i (t i )+W C i (t i ) s.t. W C i (t C i )  a[SC i (t C i ) SC i (t  i )℄ where t  i mini- mizes SC i (t) for all i. If the government implements quota, (fW C i ;q C i g i=L;H ) is the SPNE if (i) q C i maximizes W C i (q i )aSC i (q i ) (ii) q C i minimizes C i (e i (q C i ))+W C i (q i ) s.t. W C i (q C i )  a[SC i (q C i ) SC i (q  i )℄ where q  i minimizes SC i (q i ) for all i. Condition (i) says that the government sets the regulation to maximize its obje
tive fun
tion given 
ontribution s
hedule of the industry. Condition (ii) implies that the equilibrium regulation min- imizes the industry's 
ost subje
t to the parti
ipation 
onstraint of the government. Contribution s
hedules that minimize the industry's 
ost must satisfy the parti
ipation 
onstraint with equality. This means that at equilibrium, the government gets the same payoff as it gets when it does not reje
t the industry's offer. In other words, the industry has the rst mover's advantage. 63 Both 
onditions 
hara
terize the 
ontribution s
hedule. It is required that the equilibrium 
on- tribution s
hedule be W i (t i )  aSC i (t i ) aSC i (t  i ) for all t i on T when taxes are implemented. W i (q i ) aSC i (q i )aSC i (q  i ) for all q i on Q when a quota is implemented. 4.3.1 Equilibrium tax rate t C i minimizes C i (e i (t i ))+ t i e i (t i )+a[SC i (e i (t i )) SC i (t  i )℄ on T . Be
ause the best response to t i is e i (t i ) = ē i  t i =
 (if t i  
ē i ) 5 , C i (e i (t i ))+ t i e i (t i )+ [SC i (e i (t i ))SC i (t  i )℄ = 1 2  2t(ē i  t  )+ t  2 +a[ t 2  +(ē i  t  ) 2   1+ ē 2 i ℄  = 1 2  (a1)
+a  2 t 2 +2ē i  1 a   t+ a 1+ ē 2 i  : (4.2) If (a1)
+a < 0; t i = 0 is optimal 6 . If (a1)
+a  0, then t C i = 8 > < > : ē i (a
) (a1)
+a i f a > 0 otherwise: (4.3) When (a1)
+a < 0, a < 
 always holds. Hen
e, (4.3) 
hara
terizes the equilibrium tax rate. The slope of marginal damage (MD) is normalized to 1, but if the slope of the MD is d, then t C i = ē i (ad
) (a1)
+ad i f ad > 
 or t C i = 0 otherwise. Therefore, we 
an interpret a < 
 as a 
ase in that the slope of the MAC 
urve is steeper than that of a weighted MD 
urve whi
h is equal to the MD 
urve multiplied by the government's weight given to so
ial welfare. High a 
an be interpereted as high environmental awareness or low politi
al pressure from the polluting industry be
ause a lobbying group is a polluting industry only. Hen
e, we 
an interpret the above results as indi
ating that the introdu
tion of a pollution tax is blo
ked or postponed by politi
al pressure from the polluting industry when envrionmental awareness is low or politi
al 5 From F.O.C, 
(ē i  e i ) = t. And, e i (t i ) = 0 if t i > 
ē i . 6 In this 
ase, t i = 0 or t i = 
ē i might be the solution to the above problem. Be
ause C i (e i (0)) +aSC i (e i (0)) = a ē 2 i ;C i (e i (ē i ))+ 
ē i e i (
ē i )+a SC i (e i (
ē i )) = (a+1)
ē 2 i , and a < (1a)
< 
, t i = 0 is the solution. 64 pressure from the industry is high. 4.3.2 Equilibrium quota q C i minimizesC i (e i (q i ))+a[SC i (e i (q i ))SC i (q  i )℄ on Q. Be
ause the best response to q i is e i (q i ) = q i (if q i  ē i ) 7 , C i (e i (q i ))+a[SC i (e i (q i ))SC i (q  i )℄ = 1 2 [
(ē i q i ) 2 +afq 2 i + 
(ē i q i ) 2   1+ ē 2 i g℄ = 1 2 [((a+1)
+a)q 2 i 2(a+1)
ē i q i + 
(1+ 
+a
) 1+ ē 2 i ℄: (4.4) Therefore, the optimal quota is q C i = (a+1) (a+1)
+a ē i : (4.5) In order to 
ompare so
ial welfare, we 
ompare the respe
tive emission levels under both taxes and quotas. Be
ause emission in both 
ases is higher than is so
ially optimal, a lower emission level implies higher welfare. If government is benevolent (i.e.,a is innity), then taxes and quotaes are equivalent. But, if a < 
 or if e i (t C i ) q i , a (a1)
+a ē i  
(1+a) 
+a
+a ē i , 
 a 1+a ; (4.6) then the welfare under a quota is higher than the welfare under a tax. From (4.6), it is likely that a quota will be better than a tax in terms of so
ial welfare when the government pla
es mu
h more weight on 
ontributions than it does on so
ial welfare. This result is quite intuitive be
ause a tax has a negative in
ome distribution effe
t on the industry and tax payments as well as on marginal abatement 
ost. Another impli
ation of (4.6) is that a quota is always better than a tax when the slope of marginal abatement 
ost (MAC) is steeper than that of marginal damage (MD) (
 1). To 7 e i (q i ) = ē i if q i > ē i . 65 Figure 4.1: Cost savings under the tax and the quota when the slope of MAC is steeper (left) and atter (right) than the slope of MD understand this, 
onsider the 
ost savings under a tax and under a quota when an emission target in
reases by one unit. Figure 1 shows the 
ost savings under the tax and under the quota when the slope of MAC is steeper (left) and atter (right) than that of MD. The 
ost savings under the tax is area ABDC in both the left and right graphs, while 
ost savings under the quota is area CDFE. The 
ost savings under the tax is likely greater than that under the quota if the slope of MAC is steeper than that of MD. From the above effe
ts, we obtain (4.6) Be
ause a tax has a negative in
ome distribution effe
t on the industry, the industry prefers quotas over taxes. Proposition 4.1. The industry always prefers quotas over taxes under 
omplete information 
ases. Proof: See appendix C.4. 4.4 In
omplete information 
ase In the in
omplete information 
ase, we 
annot employ SPNE as an equilibrium 
on
ept; therefore, we employ Perfe
t Bayesian Equilibrium (PBE) like standard in
omplete information games. As 66 Chapter13 ofMas-
olell et al. (1995), a set of strategies and a belief fun
tion m(W)2 [0;1℄ (or m(W)2 [0;1℄) representing the government's assessment of probability that the industry is of low abatment 
ost after observing the 
ontribution s
hedule W is a PBE if (i) The industry's strategy is optimal given the government's strategies. (ii) The belief fun
tion m(W) is derived from the industry's strategy using Bayes' rule where pos- sible. 4.4.1 Tax Equation (4.3) implies that the tax rate of the low-
ost industry is weakly lower than that of the high-
ost industry. The politi
al 
ontribution under some tax rate is proportional to the differen
e in so
ial 
ost between that tax rate and the so
ially optimal tax rate. Relative to the high-
ost industry, this differen
e is small if the industry is of low 
ost. The politi
al 
ontribution of the low-
ost industry for some tax rate (lower than the so
ially optimal rate) is smaller than that of the high-
ost industry. Therefore, the high-
ost industry has an in
entive to pretend to have low 
osts. In the next sub-subse
tion, we will present the 
ase in whi
h the high-
ost industry 
annot pretend to have low 
osts (separating equilibrium). Then, we will dis
uss the 
ase in whi
h the government 
annot tell the industry's type (pooling equilibrium). 4.4.1.1 Separating equilibria Under a separating equilibrium, a government differentiates the industry's type. In other words, the high-
ost industry has no in
entive to pretend to be the low-
ost industry and in
entive 
ompatibil- ity (IC) 
onditions hold under a separating equilibrium. We 
an write IC 
onditions as C H (t C H )+ t C H e H (t C H )+W H (t C H )C H (t L )+ t L e H (t L )+W L (t L ) (4.7) C L (t L )+ t L e L (t L )+W L (t L )C L (t C H )+ t H e L (t C H )+W H (t C H ); (4.8) 67 different types have to a

ept to reveal their typeand PC 
onditions (
onditions under whi
h the government a

epts the offer and 
hooses t i for i= L;H) asW i (t i )a[SC i (t i )SC i (t  i )℄ andW i (t) W i (t i )+a[SC i (t)SC i (t i )℄ for all t =2 T and i= L;H: Let t HIC L be a tax rate su
h that (4.7) holds with equality and W L (t HIC L ) = a[SC L (t HIC L ) SC L (t  L )℄ and let be t LIC L a tax rate su
h that (4.8) holds with equality andW L (t HIC L ) =a[SC L (t HIC L )SC L (t  L )℄. t HIC L is the lowest tax rate at whi
h the high-
ost industry has no in
entive to pretend that it is of low 
ost, whereas t LIC L is the highest tax rate at whi
h the low-
ost industry has an in
entive to reveal its type 8 . Therefore, if t LIC L > t HIC L , then a separating equilibrium exists. The following proposition states the ne
essary and suf
ient 
ondition for the existen
e of a separating equilibrium. Proposition 4.2. t C H  ˆ t if and only if a separating equilibrium exists where ˆ t is a tax rate su
h that SC L ( ˆ t)SC L (t  L ) = SC H ( ˆ t)SC H (t  H ). Proof: See appendix C.5. Figures 4.2 and 4.3 give a graphi
al proof of the existen
e of a separating equilibrium. Figure 4.2 illustrates a 
ase where a separating equilibrium exists, and gure 4.3 represents a 
ase with no separating equilibrium. The PC L and PC H 
urve are a set of (t;W) that satisfy PC 
onditions for low-
ost and high-
ost industries with equality in both gures 4.2 and 4.3, respe
tively. Be
ause the government prefers high 
ontributions, the areas above PC L and PC H satises PC 
onditions for the low-
ost and high-
ost industries, respe
tively. The IC L and IC H 
urves are iso
ost 
urves for the low- and high- 
ost industries (the same 
ost as in the 
ase when the tax rate is t C H and the 
ontribution is W C H ), respe
tively. Be
use the industry likes low tax rates and small 
ontributions, the low-
ost industry prefers 
ombinations of taxes and 
ontributions in the area below IC L over (t C H ;W C H ) and the high-
ost industry prefers points below IC H over (t C H ;W C H ). Relative to the high 
ost industry, the low 
ost industry 
an endure a high tax rate be
ause it emits mu
h less. Therefore, IC L interse
ts the x-axis at a higher tax rate than IC H does. Under separating equilibria, the low 8 The low 
ost industry 
an redu
e its 
ost by de
reasing W L and in
reasing t L with IC for the high-
ost industry, holding with equality if W L (t L ) > a [SC L (t L ) SC L (t  L )℄ This is be
ause the in
rease in the sum of the abatement 
ost and the tax payment of the low-
ost industry is smaller than that of the high-
ost industry. Of 
ourse, whether W L (t L ) = a [SC L (t L )SC L (t  L )℄ is an equilibrium 
ontribution depends on the government's belief about the industry's 
ost. 68 
ost industry prefers a 
ombination of tax rate and 
ontribution over (t C H ;W C H ) but the high 
ost industry does not. In addition, the government must also a

ept that 
ombination (it prefers the 
ombination over (t C L ;W C L )). So, the 
ombination of the tax rate for the low-
ost industry and its 
ontribution under separating equilibriummust be lo
ated in the area below IC L and above IC H and PC L . The points in the grey 
olored area in gure 4.2 
an 
onstitute a separating equilibrium but no su
h area exists in gure 4.3. This is be
ause t C H  ˆ t in gure 4.2 but t C H > ˆ t in gure 4.3. As we 
an see from gure 4.2, the low 
ost industry must a

ept a higher tax rate than t C H (the tax rate of the high 
ost industry under a 
omplete information 
ase) to reveal its true type. Thus, the lower limit for the separating equilibrium tax rate of the low 
ost industry is determined by t C H . In addition, we 
an 
onsider t C H as the strength of the in
entive to pretend be
ause the high 
ost industry has this in
entive as long as the tax rate of the low 
ost industry is lower than t C H (and has a lower politi
al 
ontribution than W C H ). However, pretending to be a high 
ost industry is better for a low 
ost industry if it has to a

ept a high tax rate to reveal its type. This limit that a low 
ost industry has is an in
entive to reveal its type, and the upper limit for the separating equilibrium tax rate of the low 
ost industry is ˆ t. Be
ause of this upper limit, the low 
ost industry 
annot reveal its type if the high 
ost industry has a strong in
entive to pretend to be a low 
ost industry or if t C H is high (relative to ˆ t). Otherwise, separating equilibria are likely to exist if the pretending in
entive of the high 
ost industry is weak. We now 
onsider the set of parameters under whi
h a separating equilibrium exists more 
on
retely. When 2
ē L  ē H  ē L (ē H and ē L are not so different) and t C H > 0, ˆ t = 
(ē L + ē H )=[2(1+ 
)℄and t C H = [ē H (a 
)
℄=[(a1)
+a℄. Therefore, if ē H (a 
) (a1)
+a  
(ē L + ē H ) 2(1+ 
) , a 
+a
2 2 a 
+a ē H  ē L ; (4.9) there exists a separating equilibrium 9 . By partially differentiating (4.9) with 
 and a , we get ¶ (LHS o f (4.9)) ¶a = 2(1+ 
) 2 (a 
+a
) 2 ē H > 0; ¶ (LHS o f (4.9)) ¶ = 2
[(a 
+a
)+a℄ (a 
+a
) 2 ē H < 0: 9 We ignore a 
+a
 0. However, in this 
ase, t C H = 0, i.e. there always exists a separating equilibrium. 69 Figure 4.2: A 
ase in whi
h a tax separating equilibrium exists Figure 4.3: A 
ase in whi
h no tax separating equilibrium exists 70 Hen
e, the smaller the differen
e in natural emission (ē H  ē L ), the less the government 
ares about so
ial welfare, and/or the steeper the slope of MAC, the more likely it is that a separating equilibrium exists 10 . This result may not be intuitive, but it is quite reasonable given that the high 
ost industry has less in
entive to pretend to be the low 
ost industry, making it harder for the low 
ost industry to reveal its type. First, if the differen
e in natural emissions is smaller, then the high 
ost industry has more in
entive to pretend to have low 
osts be
ause the differen
es in tax rates and 
ontributions are smaller, i.e., the benet of pretending is smaller. Se
ond, if the government 
ares less about so
ial welfare, then the industry does not have to 
ontribute mu
h to set a tax rate that is different from the optimal one. Thus, the net benet of lobbying is large and the benet of pretending is small relative to the net benet of lobbying. Therefore, the high 
ost industry has less in
entive to pretend to be the low 
ost industry. If the slope of the MAC is steeper, the relative 
ost of lobbying to the sum of the abatement and the tax is smaller, and the tax rate of the high 
ost industry under 
omplete information is lower 11 . However, the so
ially optimal tax rates are higher, and therefore, ˆ t is high 12 . As the 
ondition for the existen
e of a separating equilibrium is t C H  ˆ t , the separating equilibrium is more likely to exist if the slope of the MAC is steeper. The next proposition des
ribes 
ases when private information improves so
ial welfare. Proposition 4.3. So
ial 
osts under a separating equilibrium are not higher than they are under 
omplete information when 
 1 and a  
. Proof: The so
ial 
osts of the high-
ost industry under the 
omplete information 
ase and separat- ing equilibrium are the same. If a  
, the so
ial 
ost of the low-
ost industry under the 
omplete information 
ases is ē 2 L =2 be
ause equation (4.3) implies that the low-
ost industry emits ē L in su
h 10 One might assume that the government 
annot differentiate the type of industry, whether low 
ost or high 
ost, ifē H = ē L ; this is true. However, the government 
an differentiate the natural emission of the industry. The industry's type does not matter be
ause what the government a
tually needs to set the tax rate is information on the natural emis- sion level of the industry. Therefore, the equilibrium under the 
ase with ē H = ē L (equivalent to one under the 
omplete information 
ase) 
an be 
onsidered a separating equilibrium in the sense that the government has the information ne
essary to set the tax rate. 11 A
tually, ¶ t C H ¶ = ē H [(a 
)
℄=[(a1)
+a ℄ 2 < 0 if t C H > 0. 12 A
tually, ˆ t = (t  H + t  L )=2, and ¶ ˆ t ¶ = (ē H +ē L ) 2 [(a 
)
℄=[1+ 
℄ 2 < 0 if t C H > 0. 71 
ases. The so
ial 
ost of the low-
ost industry under separating equilibrium is smaller than ē 2 L =2 for the following reasons; (1) the low-
ost industry redu
es the positve amount of emissions, (2) the so
ial 
ost is 
̄e 2 L =2 if the low-
ost industry does not emit at all, and (3) the so
ial 
ost de- 
reases with emissions down to the so
ially optimal emission level, then in
reases until there are no emissions. QED As mentioned in the proof of propostion 3, equation (4.3) states that tax rates are zero under 
om- plete information 
ases if a  
. Therefore, the industry does not redu
e its emissions at all under su
h 
ases. However, be
ause tax rate for the low-
ost industry is positive under separating equilib- ria, the low-
ost industry redu
es its emissions but it may redu
e too mu
h. The so
ial 
ost in the 
ase of ex
essive redu
tion is at most 
ē 2 L =2 (= so
ial 
ost under the no-emissions 
ase). If 
  1, 
ē 2 L =2< ē 2 L =2. Therefore, private information might improve so
ial welfare under taxes. 4.4.1.2 Pooling equilibria Under pooling equilibria, the government 
annot differentiate the types of the industries. In other words, the high 
ost industry has an in
entive to prevent the government from differentiating it under pooling equilibira. Te
hni
ally, any t, 0  t  
ē H 
an be a pooling equilibrium rate su
h that 1 2 F(ē H )C H (e H (t))+ te H (t)+W P (t): (4.10) where W P (t) aE ē [SC(t)SC(t P )℄ and t P = argminE ē [SC(t)℄. t must satisfyW P (t 0 )W P (t)+ aE ē [SC(t 0 ) SC i (t)℄ for any t 0 6= t. Let t P 1 and t P 2 be a smaller t and bigger t, respe
tively, that satisfy (4.10) with equality 13 . Although the total 
ost of the high-
ost industry under pooling equilibria is not higher than it is under the 
omplete information 
ase or separating equilibria, it is un
lear whether the total 
ost of the low-
ost industry is higher than it is under separating equilibrium. Whether the total 
ost of the low-
ost industry is higher than that under separating equilibrium depends on the pooling 13 If t P 1 < 0, t P 1 = 0: If t P 2 > 
ē H , t P 2 = 
ē H . 72 equilibrium tax rate, i.e., the government's belief about the industry's type and parameters. In order to 
ompare the results under separating equilibrium to those under pooling equilibrium, in the next se
tion we fo
us on the equilibria where the total 
ost of the low-
ost industry is the lowest. We 
an eliminate pooling equilibria by domination-based renement of beliefs if the total 
ost of the low-
ost industry under the best separating equilibrium is lower than the total 
ost under pooling equilibria. Be
ause the total 
ost of the industry at the given tax rate is minimized when W P (t)= aE ē [SC(t)SC(t P )℄, the tax rate at whi
h the 
ost of the low 
ost industry is minimized 14 is t P = argmin t0 te L (t)+C L (e L (t))+W P (t) = 1 2  (a1)
+a  2 t 2 +2  ē L  a  E[ē℄  t+ a 1+ E[ē L ℄ 2  : (4.11) If (a1)
+a < 0 or E[ē℄a < 
ē L 15 , t P = 0. Otherwise, t P = 8 > < > : t P 2 i f t p 2 < (E[ē℄a
ē L ) (a1)
+a (E[ē℄a
ē L ) (a1)
+a otherwise: (4.12) Again, this t P will be used in the numeri
al example. 4.4.2 Quotas The low-
ost industry might have an in
entive to pretend that it has high 
osts be
ause it would benet from a large quota. Besides, equation (4.5) implies that the high-
ost industry is given a larger quota than the low-
ost industry under the 
omplete information 
ase. On the other hand, the high-
ost industry does not have an in
entive to pretend that it has low 
osts. Therefore, in the next sub-subse
tion, we will des
ribe the 
ase in whi
h the low-
ost industry has no in
entive to pretend to have high 
osts (separating equilibrium). Then, we will dis
uss the 
ase in whi
h the 14 We assume that (1+ 
)ē L  E[ē℄. This assumption ensures that t P = 
E[ē℄=(1+ 
) and (E[ē℄a
ē L ) (a1)
+a < 
ē L . See appendix C.1 when (1+ 
)ē L < E[ē℄. 15 In this 
ase, t P = 0 satises Eq(4.10) 73 government 
annot tell the type of the industry (pooling equilibrium). 4.4.2.1 Separating equilibria The 
onditions for a separating equilibrium are the IC 
onditions (Eq (4.13) for the low-
ost indus- try and (4.14) 16 for the high-
ost industry),different types have to a

ept to reveal their type C L (q C L )+W C L (q C L )C L (e L (q H ))+W C H (q H ) (4.13) C H (q C L )+W C L (q C L )C H (q H )+W C H (q H ) (4.14) and PC 
onditions, W i (q i )  a[SC i (q i ) SC i (q  i )℄ for i = L;H and W i (q)  W i (q i )+a[SC i (q) SC i (q i )℄ for q 2 Q and i= L;H. q H must be larger than q  H . Unlike taxes, a separating equilibrium always exists under quotas. Proposition 4.4. There exists a separating equilibrium under quotas. Proof: See appendix C.6. The reason why a separating equilibrium always exists under quotas is as follows. The low-
ost industry wants a large quota less than the high-
ost industry does. Therefore, when the high- 
ost industry puts pressure on the government to set a larger quota, the low-
ost industry has less in
entive to pretend. Finally, the low-
ost industry has no in
entive to enlarge the quota. When a tax is implemented, the low-
ost industry reveals its type by making the government set the tax rate for the low-
ost industry higher than that for the high-
ost industry. However, the tax rate for the high-
ost industry under the 
omplete information 
ase is greater than that for the low-
ost industry, so it is sometimes hard for the low-
ost industry to make the government set the tax rate higher for the low-
ost industry. This is why a tax separating equilibrium might not exist. On the other hand, the quota for the high-
ost industry under 
omplete information is larger than that for the low-
ost industry, and the high-
ost industry more strongly advo
ates for a large quota than the low-
ost industry. Therefore, it is not dif
ult for the high-
ost industry to make the government set the quota high enough that the low-
ost industry will have no in
entive to pretend. 16 If ē L  q H , then this 
ondition is the same as C L (q C L )+W C L (q C L )W C H (q H ). 74 As for so
ial welfare, the so
ial 
ost under separating equilibria is not lower than that under 
om- plete information. The high-
ost industry might make the government set the quota higher than it would under the 
omplete information 
ase in order to reveal its type 17 . The low-
ost industry re
eives the same quota as it does in the 
omplete information 
ase. Thus, private information does not improve so
ial welfare under quotas. 4.4.2.2 Pooling equilibria Whether a separating equilibrium is a
hieved depends on the government's belief about the indus- try's type, even if it always exists. Therefore, it is possible that pooling equilibria also exist. If a pooling equilibrium exists, the low-
ost industry has an in
entive not to make the government differentiate its type. In this sub-subse
tion, we investigate the property of pooling equilibria. Te
hni
ally, any q 
an be a pooling equilibrium su
h that C L (q C L )+W C L (q C L )C L (e L (q))+W P (q) (4.15) where W P (q)  a[E ē [SC(q) SC(q  )℄℄ and q  = argmin E ē [SC(q)℄. q must satisfy W P (q 0 )  W P (q) + aE ē [SC(q 0 ) SC i (q)℄ for any q 0 6= q. Let q P 1 and q P 2 be a smaller q and bigger q, re- spe
tively, that satisfy (4.15) with equality 18 . While the total 
ost of the low-
ost industry under pooling equilibria is not higher than that under the 
omplete information 
ase or separating equilibria, it is ambiguous whether the total 
ost of the high-
ost industry is higher than it is under separating equilibrium. Whether the total 
ost of the high-
ost industry is higher than that under separating equilibrium depends on the pooling equilibrium quota, i.e., the government's belief about ea
h industry's type and parameters. In order to 
ompare the results obtained under separating equilibrium to those obtained under pooling equilibrium, in the next se
tion we fo
us on the equilibrium where the total 
ost of the high-
ost industry is the lowest. Be
ause the total 
ost of the industry at the given quota is minimized when W P (q) = aE ē [SC(q) SC(q P )℄, the quota at whi
h the 
ost of the high-
ost industry is 17 Depending on parameters, the low-
ost industry does not have an in
entive to pretend to be a high-
ost industry. 18 If q P 1 < 0, q P 1 = 0. If q P 2 > ē H , q P 2 = ē H . 75 minimized 19 is q P H = arg min q P 1 qq P 2 C H (q)+W P (q) = 1 2  f
+(1+ 
)agq 2 2(
ē H +a
E[ē℄)q+ 
ē 2 H + a 2 1+ E[ē℄ 2  : (4.16) From rst order 
onditions (F.O.Cs), q P H = 8 > < > : a
E[ē℄+
ē H 
+(1+
)a i f q P 2 < a
E[ē℄+
ē H 
+(1+
)a q P 2 otherwise: (4.17) 4.4.3 Differen
e in results and in
entives between tax and quota In this subse
tion, we briey and intuitively review what happens under tax and quota. In the last subse
tion, we showed that private information may worsen so
ial welfare under quota but may improve it under tax. The type that has the in
entive to pretend to be the other is different between tax and quota. In addition, the dire
tion of 
hange in the regulation level depends on whi
h type has to reveal its type (or whi
h type has an in
entive to pretend to be the other type). These fa
tors result in the differen
es under tax and quota. The high 
ost industry more strongly desires lax regulation or has a higher willingness to pay than the low 
ost industry. Roughly speaking, the high 
ost industry prefers the 
ombination of the high politi
al 
ontribution and lax regulation to that of the low politi
al 
ontribution and stringent regula- tion, while the low 
ost industry prefers the 
ombination of low politi
al 
ontribution and stringent regulation. Therefore, the high 
ost industry 
an reveal its type by a

epting laxer regulations and higher politi
al 
ontributions relative to the 
omplete information 
ase be
ause benet. If the high 
ost industry does so, so
ial welfare de
reases relative to the 
omplete information 
ase. Thus, private information worsens so
ial welfare if the high 
ost industry has to reveal its type. However, the low 
ost industry 
annot reveal its type by a

epting laxer regulations and higher politi
al 
on- tributions relative to the 
omplete information 
ase. If the low 
ost industry has to reveal its type, it must also a

ept stringent regulations by paying low politi
al 
ontributions relative to the 
omplete 19 We assume that (1+ 
)ē L  
E[ē℄ and ē L  q. See appendix C.2 about other 
ases. 76 information 
ase, whi
h means that private information may improve so
ial welfare when the low 
ost industry has to reveal its type be
ause regulations under the 
omplete information 
ase are too lax due to the lobbying of the industry. The industry prefers a high quota and low tax rate regardless of type. Under the 
omplete informa- tion 
ase, the quota of the high 
ost industry is higher than that of the low 
ost industry, and the tax rate of the low 
ost industry is lower. Under quota, the low 
ost industry might have an in
entive to pretend to be high 
ost, or the high 
ost industry might have to reveal its type. Therefore, private information may worsen so
ial welfare under quota. Under tax, however, private information may improve so
ial welfare be
ause the low 
ost industry may be for
ed to reveal its type. Be
ause of the two differen
es above, we obtain different results of the separating equilibrium under quota and tax. Private information may improve so
ial welfare under tax but may worsen so
ial welfare under quota. As we show in the last se
tion, quota is generally better than tax under the 
omplete information 
ase due to the negative in
ome distribution effe
t on the industry. In the next se
tion, we examine how private information inuen
es the dominan
e of quota over tax using numeri
al examples. 4.4.4 Renements of beliefs Whi
h set of tax rates or quotas and politi
al 
ontribution s
hedules 
onstitutes an equilibrium de- pends on the government's belief about the industry's type. The simplest renement, the domination- based renement of beliefs des
ribed in the appendix of Chapter 13 of Mas-
olell et al. (1995), is based on the idea that reasonable beliefs should not assign a positive probability to an industry's taking an a
tion that is stri
tly dominated for that industry. This renement eliminates all sepa- rating equilibria ex
ept for the best one. However, it 
annot rule out pooling equilibria, where the payoff of the low type is higher than it is in the best separating equilibrium under taxes and where the high type's payoff is higher than it is in the best separating equilibrium under quotas. Be
ause the total 
osts of the industry under separating and pooling equilibria are 
ompli
ated, we will ex- amine in the next se
tion whether pooling equilibria are eliminated by the renement of beliefs by using numeri
al examples. 77 4.5 Numeri
al examples Figures 4.4, 4.5, and 4.6 show so
ial 
osts of the high-
ost industry (gures 4.4(a), 4.5(a), and 4.6(a)) and the low-
ost industry (4.4(b), 4.5(b), and 4.6(b)), the industries' 
osts for the high-
ost industry (gures 4.4(
), 4.5(
), and 4.6(
)) and the low-
ost industry (4.4(d), 4.5(d), and 4.6(d)), and politi
al 
ontributions (gures 4.4(e), 4.5(e), and 4.6(e)) under different equilibria with 
hanging a when 
 is 1=3, 1 or 3. The natural emission levels of the high- and low-
ost industry are 80 and 50, respe
tively. The politi
al 
ontribution of the industry under taxes and 
omplete information is in
reasing with a until a = 
, the maximum value of a where the equilibrium tax rates are zero and de
reasing after that. The 
ontribution of the low-
ost industry under the tax separating equilibrium is in
reasing with a when a is very small relative to 
, de
reasing until the value of a is 
lose to 
, and then in
reasing. In addition, no separating equilibrium exists when a is greater than the value at whi
h the 
ontribution of the low-
ost industry under tax separating equilibrium is 
lose to that of the high- 
ost industry under taxes. The politi
al 
ontribution under quotas with 
omplete information is in
reasing with a when it is very small. Otherwise, the 
ontribution is de
reasing. The 
ontribution of the high abatement 
ost industry under the separating equilibrium (the same as the 
omplete information 
ase) is in
reasing in a mu
h wider range of a . When taxes are implemented, so
ial welfare under separating equilibrium is higher than it is under the 
omplete information 
ase if the industry is of low abatement 
ost and a is less than two times greater than 
 (gures 4.4(b), 4.5(b) and 4.6(b)). Hen
e, private information improves so
ial wel- fare when the government has little 
on
ern for so
ial welfare. The industry's 
ost under pooling equilibria (best for the low-
ost industry) is smaller than it is under the separating equilibrium in all 
ases (
= 1=3, 1, and 3). Therefore, we 
annot eliminate pooling equilibria. The industry's 
ost is de
reasing in p L , the prior belief that the industry is of low 
ost. Under quotas, the low-
ost industry has more in
entive to pretend to be a high-
ost industry when the slope of marginal abatement 
ost (MAC) is higher (
 is higher) and the government 
ares so
ial welfare more (i.e., a is greater). When the government 
ares more about so
ial welfare, the industry has to pay a larger politi
al 
ontribution to get a larger quota. Thus, pretending to be a high-
ost 78 (a) So
ial 
osts of the high-
ost industy (b) So
ial 
osts of the low-
ost industy (
) High-
ost industy' 
osts (d) Low-
ost industy' 
osts (e) Politi
al 
ontributions Figure 4.4: So
ial 
osts, industries' 
osts, and politi
al 
ontributions when 
= 1=3 79 (a) So
ial 
osts of the high-
ost industy (b) So
ial 
osts of the low-
ost industy (
) High-
ost industy's 
osts (d) Low-
ost industy's 
osts (e) Politi
al 
ontributions Figure 4.5: So
ial 
osts, industries' 
osts, and politi
al 
ontributions when 
= 1 80 (a) So
ial 
osts of the high-
ost industy (b) So
ial 
osts of the low-
ost industy (
) High 
ost industy's 
ost (d) Low 
ost industy's 
ost (e) Politi
al 
ontributions Figure 4.6: So
ial 
osts, industries' 
osts, and politi
al 
ontributions when 
= 3 81 industry is a more attra
tive way for the low-
ost industry to get a larger quota. When the slope of MAC is at (
= 1=3 and 1), the high-
ost industry's 
ost under the best separating equilibrium is smaller than it is under the pooling equilibrium. However, the opposite happens when the slope of MAC is steep (
 = 3). The so
ially optimal quota under the pooling equilibrium is the same as that for the high-
ost industry when the slope of MAC is steep and there is only a slight prior belief that the industry is of low 
ost. Under su
h a 
ase, the industry 
an obtain the same quota with a smaller politi
al 
ontribution under a pooling equilibrium than it 
an under the separating equilibrium. This is why the high-
ost industry's 
ost under the pooling equilibrium is smaller than it is under the separating equilibrium with the steep MAC 
urve (
= 3). For 
omparison of the results under taxes with those under quotas (gures 4.4(
), 4.4(d), 4.5(
), 4.5(d), 4.6(
), and 4.6(d)), the industry seems to almost always prefer quotas over taxes under any type of equilibrium. This is quite intuitive be
ause the industry must pay the tax as well as the abatement expense under taxes, whereas it pays only the abatement expense under quotas. Under the 
omplete information 
ase, the quota is always better than the tax if 
  1 20 . under the in
omplete information 
ase, sometimes (i.e., when the value of a is similar to that of 
) the tax is better than the quota when the industry is of low 
ost. Figure 4.7 shows the lowest separating equilibrium quota for the high abatement 
ost industry under the 
ases where the low abatement 
ost industry has different natural emission levels (ē L = 50, 60, and 70) with 
hanging a , whereas gure 4.8 shows the lowest separating equilibrium tax rate for the low abatement 
ost industry under the 
ases where the high abatement 
ost industry has different natural emission levels (ē H = 60, 70, and 80) with 
hanging a . The high abatement 
ost industry with higher natural emission has more in
entive to pretend to be the low abatement 
ost industry (i.e., the separating equilibrium tax rate is high) be
ause the differen
e in the politi
al 
ontributions ne
essary to set the same (low) tax rate between the low- and high-
ost industries is larger when a high-
ost industry has more natural emission. In other words, the high-
ost industry with higher natural emission will get more benets from pretending to be a low-
ost industry if taxes are implemented. If quotas are implemented, the lowest separating equilibrium quota for the low-
ost industry with 20 Eq (4.6) always holds if 
 1. 82 a lower natural emission level is smaller when a is small. This is be
ause the quota for the high- 
ost industry under the 
omplete information 
ase is larger than the natural emission level of the low-
ost industry, so the low-
ost industry with a lower natural emission level has less in
entive to pretend to be a high-
ost industry. On the other hand, the lowest separating equilibrium quota for the low-
ost industry with a lower natural emission level is smaller when a is large. In this 
ase, the quota for the high-
ost industry under the 
omplete information 
ase is smaller than the natural emission level of the low-
ost industry. In addition, the smaller the natural emission level of the low-
ost industry is, the larger the differen
e in the quotas under 
omplete information between the high- and low-
ost industries be
omes. Therefore, the low-
ost industry with a lower natural emission level has more in
entive to pretend to be a high-
ost industry. 4.6 Con
lusion We 
ompare taxes with quotas and 
omplete information 
ases with in
omplete information 
ases when levels of regulation are affe
ted by an industry lobby. We show that quotas are so
ially preferred over taxes when the government 
ares little about so
ial benet very mu
h or when the slope of MAC is steeper than that of MD. However, private information might redu
e the 
omparative disadvantage of taxes 
ompared to quotas. This result is totally different from the result in the literature on tax versus quota. If the slope of the MAC 
urve is steep relative to that of the MD 
urve, the industry's benets from lobbying are larger under a tax than they are under a quota, so the industry lobbies more strongly against taxes than against quotas. The impa
t on so
ial welfare of this s
enario is greater than that of asymmetri
 information. Therefore, we might be able to say that politi
al e
onomy issues indu
e greater distortion than asymmetri
 information does. We also show that the tax rate for the low abatement 
ost industry is higher than that for the high abatement 
ost industry if a separating equilibrium exists. This result is 
onsistent with the demonstration byEkins and Spe
k (1999) that the effe
tive tax rates for emission-intensive or manufa
turing industries are mu
h lower than the nominal tax rates for other se
tors in Sweden, Denmark, and Norway. Both pro-environmental poli
y groups and polluting industries lobby on environmental poli
ies, 83 Figure 4.7: Separating equilibrium quotas for the high-
ost industry with the low-
ost indsutry with different levels of natural emissions(ē H = 60,70,and 80) Figure 4.8: Separating equilibrium tax rate for the low-
ost industry with high-
ost indsutry with different levels of natural emissions(ē H = 60,70,and 80) 84 although the former's 
ontributions are very small 
ompared to those of the latter. Hen
e, extending themodel of this 
hapter to lobbying bymultiple prin
ipals with private information is one dire
tion of future resear
h. 85 Chapter 5 Con
lusion 5.1 Summary of 
ontributions This thesis theoreti
ally and empiri
ally examined environmental poli
ies, in
luding voluntary ap- proa
hes, quotas and market-based instruments. All of these poli
ies have been effe
tive as al- ternatives to traditional environmental regulation approa
hes su
h as the 
ommand and 
ontrol approa
h. In our theoreti
al analysis of voluntary emissions redu
tion programs, we built a model that ex- plains why governments implement voluntary programs even though the parti
ipation rates are sometimes low. This reason has not been explained by the literature on voluntary approa
hes to environmental prote
tion. Only Dawson and Segerson (2008)analyzed voluntary programs while in
orporating individual rms' parti
ipation de
ision into their model. In their model, the program generates lower so
ial welfare than the mandatory poli
y. On the other hand, our model implies that the voluntary program 
an generate greater so
ial welfare and aggregate abatement than a mandatory standard if it is politi
ally dif
ult to set a satisfa
tory standard. Therefore, regulators sometime implement a voluntary program in su
h a 
ase. In this situation, the regulator sets the abatement rate of the program as high as possible, subje
t to the 
onstraint that the abatement 
ost of parti
ipating rms is smaller than their 
ost under the mandatory standard. In addition to 
on- straints on the 
ost of individual parti
ipating rms, no more rms will parti
ipate in the program if the aggregate abatement by the rms already parti
ipating is greater than under the mandatory standard. Our model implies that the parti
ipation rate of the program will be low if the mandatory standard is set very lax. 86 We empiri
ally investigated the determinants of ISO 14001 adoption and its impa
ts on environ- mental performan
e in 
hapter 3. There are numerous studies on them, but none has addressed intra-industry spillovers of ISO 14001 adoption and environmental performan
e. We examined su
h spillovers by applying spatial e
onometri
 estimation methods to a Japanese fa
ility dataset. We found that fa
ilities emitting into water are more likely to adopt ISO 14001 if fa
ilities that belong to rms with similar revenue in the same industry adopt this standard. We also found that the per
entage 
hange in emissions into air for similarly sized fa
ilities in the same industry are 
orrelated. The estimated spillover effe
ts of ISO 14001 adoption between similarly sized rms emitting into water are not small. Thus, our result implies that the fa
t that rival rms adopt ISO 14001 is an important motive for its adoption. In 
hapter 4, we examined the effe
ts of private information about emission abatement 
ost on lobbying a
tivity of a polluting industry and on so
ial welfare under taxes and quotas. We showed that to reveal its abatement 
ost information, an industry with a low abatement 
ost makes to the government set a higher tax rate than a tax rate set under 
omplete information that is too low in terms of the so
ial welfare. Thus, private information might improve the so
ial welfare under taxes, in 
ontrast to models with a benevolent government. On the other hand, under quotas, an industry with a high abatement 
ost might have to make the government set a larger quota than a quota set under 
omplete information 
ase so that it reveals 
ost information. Thus, private information does not improve so
ial welfare. In this 
hapter, we also 
ompared taxes and quotas and showed that quotas are so
ially preferred when the slope of marginal abatement 
ost is steeper than that of the marginal damage or when the government weighs politi
al 
ontribution mu
h more than so
ial welfare. 5.2 Dire
tions for further resear
h We 
on
lude this thesis by noting unaddressed issues to be studied in future resear
h. The major unaddressed issue on voluntary approa
hes related to environmental issues is information dis
lo- sure. Both theoreti
al and empiri
al analyses of voluntary approa
hes deal with voluntary a
tions su
h as improvement of environmental performan
e and a
quirement of environmental manage- 87 ment system 
erti
ates. However, many rms voluntarily provide self-reported information on pollution levels in their Corporate So
ial Responsible report, environmental reports, and/or web- site. Similar to the Carbon Dis
losure Proje
t (CDP), Prin
iples for Responsible Investment (PRI), and the Investor Network on Climate Risk, we have worldwide shareholder's initiatives that require rms to dis
lose information on their greenhouse gas (GHG) emissions. Thus, voluntary informa- tion dis
losure has been pervasive, and I believe that it is meaningful to examine determinants and impa
ts of voluntary information dis
losure. In parti
ular, it is interesting to examine the relationship between rms' voluntary information dis
losure and their e
onomi
 and environmental performan
e, but we have to examine this rela- tionship with great 
are. For example, rms are more likely to voluntarily dis
lose information on their environmental performan
e if their environmental performan
e is good. When information is available to the publi
, rms are subje
t to higher pressure to redu
e emissions and a
tually re- du
e emissions. Thus, we have to 
ontrol for dual 
ausality between information dis
losure and environmental performan
e. The 
redibility of self-reported information is also worth resear
hing. For example, does third party 
erti
ation on self-reported information have impa
ts on a rm's e
onomi
 performan
e or its sto
k pri
e? How do we make rms report truthfully without mu
h 
ost? Su
h questions remain for to be answered by future resear
h. Due to the limitations of the approa
hes we used in this thesis, there are issues that remain to be addressed. In our analysis of voluntary pollutant redu
tion programs, we emphasized the politi
al dif
ulty of setting mandatory poli
ies. Heterogeneity of polluting rms is also likely to play an important role in the implementation of voluntary programs, but rms were homogeneous in our model. Introdu
tion of the heterogeneity makes a model mu
h more 
ompli
ated, and therefore, we would have had to 
al
ulate and 
he
k many potential equilibria for the implementation of a voluntary program. In our empiri
al analysis, we did not determine the fa
tors that indu
ed the intra-industry spillovers of ISO 14001 adoption and environmental performan
e that we found. Therefore, identifying the 
auses of the intra-industry spillovers 
an be one dire
tion of future resear
h. Our model of tax versus quota analysis involved only one industry lobby group, but a model with 
ompeting lobby groups would be more realisti
 and might have mu
h more relevant impli
ations. 88 Bibliography Aidt, Toke S. (1998) “Politi
al Internalization of E
onomi
 Externalities and Environmental Pol- i
y,” Journal of Publi
 E
onomi
s, Vol. 69, No. 1, pp. 1-16. Antweiler, Werner and Kathryn Harrison (2007) “Canada's Voluntary ARET Program: Limited Su

ess Despite Industry Cosponsorship,” Journal of Poli
y Analysis and Management, Vol. 26, No. 4, pp. 755-774. Arimura, Toshi Hide, Akira Hibiki, and Hajime Katayama (2008) “Is a voluntary approa
h an effe
tive environmental poli
y instrument?: A 
ase for environmental management systems,” Journal of Environmental E
onomi
s and Management, Vol. 55, No. 3, pp. 281-295. Arimura, Toshi Hide, Ni
ole Darnall, and Hajime Katayama (2009) “Is ISO 14001 a Gateway to More Advan
ed Voluntary A
tion? A Case for Green Supply Chain Management,” Dis
ussion Papers dp-09-05, Resour
es For the Future. Austen-Smith, David (1995) “Campaign 
ontributions and a

ess,” Ameri
an Politi
al S
ien
e Re- view, Vol. 89, No. 3, pp. 566-581. Austen-Smith, David and Jeffrey S. Banks (2002) “Costly signaling and 
heap talk in models of politi
al inuen
e,” European Journal of Politi
al E
onomy, Vol. 18, No. 2, pp. 263-280. Austen-Smith, David and John Wright (1994) “Countera
tive Lobbying,” Ameri
an Journal of Politi
al S
ien
e, Vol. 38, No. 1, pp. 25-44. Barbier, Edward B., Ri
hard Damania, and Daniel Leonard (2005) “Corruption, trade and resour
e 
onversion,” Journal of Environmental E
onomi
s andManagement, Vol. 50, No. 2, pp. 276-299. Barla, Philippe (2007) “ISO 14001 
erti
ation and environmental performan
e in Quebe
's pulp and paper industry,” Journal of Environmental E
onomi
s and Management, Vol. 53, No. 3, pp. 291-306. 89 Barrett, S
ott (1994) “Self-enfor
ing International Environmental Agreements,” Oxford E
onomi Papers, Vol. 46, pp. 878-894. Bennedsen, Morten and Sven E. Feldmann (2006) “Informational lobbying and politi
al 
ontribu- tions,” Journal of Publi
 E
onomi
s, Vol. 90, No. 4-5, pp. 631-656. Berenbeim, Ronald E. (1992) The Corporate Ethi
s Pra
ti
es, New York: Conferen
e Board. Boyer, Mar
el and Jean-Ja
ques Laffont (1999) “Toward a politi
al theory of the emergen
e of environmental in
entive regulation,” RAND Journal of E
onomi
s, Vol. 30, No. 1, pp. 276-299. Brouhle, Keith, Charles Grifths, and Ann Wolverton (2005) “The Use of Voluntary Approa
hes for Environmental Poli
ymaking in the U.S,” in Cro
i, Edoardo ed. The Handbook of Environ- mental Voluntary Agreements, Netherlands: Springer. Con
oni, Paola (2003) “Green lobbies and transboundary pollution in large open e
onomies,” Jour- nal of International E
onomi
s, Vol. 102, No. 59, pp. 399-422. Costello, Christopher and Larry Karp (2005) “Dynami
 taxes and quotas with learning,” Journal of E
onomi
 Dynami
s and Control, Vol. 28, No. 8, pp. 1661-1680. Dahm, Matthias and Ni
olás Porteiro (2008) “Informational lobbying under the shadow of politi
al pressure,” So
ial Choi
e Welfare, Vol. 30, No. 4, pp. 531-559. Damania, Ri
hard (2001) “When the Weak Win: The Role of Investment in Environmental Lobby- ing,” Journal of Environmental E
onomi
s and Management, Vol. 42, No. 1, pp. 1-22. Darnall, Ni
ole and Daniel Edwards (2006) “Predi
ting the 
ost of environmental management system adoption: The role of 
apabilities, resour
es and ownership stru
ture,” Strategi
 Man- agement Journal, Vol. 27, No. 2, pp. 301-320. Darnall, Ni
ole and Alexei Pavli
hev (2004) “Environmental poli
y tools and rm-level manage- ment in the United States.” available at: http://www.oe
d.org/dataoe
d/1/59/35590060.pdf. Darnall, Ni
ole and Stephen Side (2008) “Assessing the performan
e of voluntary environmental programs: Does 
erti
ation matter?” Poli
y Studies Journal, Vol. 36, No. 1, pp. 95-117. 90 Dasgupta, Partha, Peter Hammond, and Eri
 Maskin (1980) “On Imperfe
t Information and Opti- mal Pollution Control,” Review of E
onomi
 Studies, Vol. 47, No. 5, pp. 857-860. Dawson, Na Li and Kathleen Segerson (2008) “Voluntary Agreements with Industries: Parti
ipa- tion In
entives with Industry-Wide Targets,” Land E
onomi
s, Vol. 84, No. 1, pp. 97-114. Ekins, Paul and Stefan Spe
k (1999) “Competitiveness and Exemptions From Environmental Taxes in Europe,” Environmental and Resour
e E
onomi
s. European Environment Agen
y (1997) Environmental Agreements: Environmental Effe
tiveness: Environmental Issue Series No.3 1-2. Fredriksson, Per G. (1997) “The politi
al e
onomy of pollution taxes in a small open e
onomy,” Journal of Environmental E
onomi
s and Management, Vol. 33, No. 1, pp. 44-58. Gamper-Rabindran, Shanti (2006) “Did the EPA's Voluntary Industrial Toxi
s Program Redu
e Emission? A GIS Analysis of Distributional Impa
ts and By-media Analysis of Substitution,” Journal of Environmental E
onomi
s and Management, Vol. 52, No. 1, pp. 391-410. Gla
hant, Matthieu (2007) “Non-binding Voluntary Agreements,” Journal of Environmental E
o- nomi
s and Management, Vol. 32-48, No. 1, p. 54. Gla
hant, Matthieu, Marta Vi
arelli, and Fr'ed'erique Vin
ent (2004) “Management Environ- nemental : Une Enquete aupres de 270 Etablissements Industriels Fran
ais.” available at: http://www.oe
d.org/dataoe
d/25/37/31684454.pdf. Grossman, Gene and Elhanan Helpman (1994) “Menu au
tions, resour
e all
ation, and e
onomi inuen
e,” Ameri
an E
onomi
 Review, Vol. 84, No. 4, pp. 833-850. Hansen, Lars. G. (1999) “Environmental Regulation through Voluntary Agreements,” in Car- raro, Carlo and François Lévêque eds. Voluntary Approa
hes in Environmental Poli
y, Boston: Kluwer A
ademi
. 91 Hepple, Leslie W. (2004) “Bayesian Model Choi
e in Spatial E
onometri
s,” in Fomby, Thomas B. and R. Carter Hill eds. Spatial and Spatiotemporal E
onometri
s (Advan
es in E
onometri
s, Volume 18), UK: Emerald Group Publishing Limited. Hibiki, Akira and Toshi Hide Arimura (2004) “Environmental poli
y tools and rm-level manage- ment in Japan.” available at: http://www.oe
d.org/dataoe
d/26/0/31686250.pdf. Hoel, Mi
hael and Larry Karp (2001) “Taxes and quotas for a sto
k pollutant with multipli
ative un
ertainty,” Journal of Publi
 E
onomi
s, Vol. 82, No. 1, pp. 91-114. (2002) “Taxes versus quotas for a sto
k pollutant,” Resour
e and Energy E
onomi
s, Vol. 24, No. 4, pp. 367-384. Innes, Robert and Abdoul G. Sam (2008) “Voluntary Pollution Redu
tions and the Enfor
ement of Environmental Law: An Empiri
al Study of the 33/50 Program,” Journal of Law and E
onomi
s, Vol. 51, No. 2, pp. 271-296. Kaplow, Louis and Steven Shavell (2002) “On the Superiority of Corre
tive Taxes to Quantity Regulation,” Ameri
an Law and E
onomi
s Review, Vol. 4, No. 1, pp. 1-17. Karp, Larry S. and Jiangfeng Zhang (2002) “Controlling a Sto
k Pollutant with Endogenous Invest- ment and Asymmetri
 Information,” working papers, Department of Agri
ultural and Resour
e E
onomi
s, University of California, Berkeley. Kerekes, Sándor, Gábor Harangozó, Patrí
ia Németh, and Á. Nem
si
sné Zsóka (2004) “Environmental poli
y tools and rm-level management in Hungary.” available at: http://www.oe
d.org/dataoe
d/26/0/31686250.pdf. Khanna, Madhu (2007) “The U.S. 33/50 Voluntary Program: Its Design and Effe
tiveness,” in Mor- genstern, Ri
hard D. and William Aaron Pizer eds. Reality Che
k the Nature and Performan
e of Voluntary Environmental programs in the United States, Europe, and Japan, Washington DC: Resour
es for the Future. 92 King, Andrew A., Mi
hael J. Lenox, and Ann Terlaak (2005) “The strategi
 use of de
entralized institutions: Exploring 
erti
ation with the ISO 14001 management standard,” Journal of Envi- ronmental E
onomi
s and Management, Vol. 48, No. 6, pp. 1091-1106. LeSage, James and R. Kelly Pa
e (2009) Introdu
tion to Spatial E
onometri
s, Bo
a Raton, FL: CRC Press. Lewis, Tra
y R. (1996) “Prote
ting the Environment When Costs and Benets are Privately Known,” RAND journal of E
onomi
s, Vol. 27, No. 4, pp. 819-847. Lohmann, Suzanne (1995) “Information, A

ess, and Contributions: A Signaling Model of Lob- bying,” Publi
 Choi
e, Vol. 85, No. 3-4, pp. 267-284. Lutz, Stefan, Thomas P. Lyon, and John W. Maxwell (2000) “Quality leadership when regulatory standards are forth
oming,” Journal of Industrial E
onomi
s, Vol. 48, No. 3, pp. 331-348. Lyon, Thomas P. and John W. Maxwell (2002) “Voluntary Approa
hes to Environmental Regu- lation: A Survey,” in Franzini, Maurizio and Antonio Ni
ita eds. E
onomi
 Institutions and Environmental Poli
y: Post Present and Future, Hampshire, UK: Ashgate Publishing Ltd. (2003) “Self-regulation, Taxation and Publi
 Voluntary Environmental Agreements,” Journal of Publi
 E
onomi
s, Vol. 87, No. 3-4, pp. 1453-1486. (2004) Corporate Environmentalism and Publi
 Poli
y, Cambridge, UK: Cambridge Uni- versity Press. Manzini, Paola and Mar
o Mariotti (2003) “A Bargaining Model of voluntary Environmental Agreements,” Journal of Publi
 E
onomi
s, Vol. 87, No. 12, pp. 2725-2736. Mas-
olell, Andrew, Mi
hael D. Whinston, and Jerry R. Green (1995) Mi
roe
onomi
 Theory, New York: Oxford University Press. Maskin, Eri
 and Jean Tirole (1992) “The prin
ipal-agent relationship with an informed prin
ipal, II: Common values,” E
onometria, Vol. 60, No. 1, pp. 1-42. 93 Maxwell, John W., Thomas P. Lyon, and Steven C. Ha
kett (2000) “Self-Regulation and So
ial Welfare: The Politi
al E
onomy of Corporate Environmentalism,” Journal of Law and E
o- nomi
s, Vol. 43, No. 3, pp. 583-618. Melnyk, Steven A., Robert P. Sroufe, and Roger Calantone (2003) “Assessing the impa
t of envi- ronmental management systems on 
orporate and environmental performan
e,” Journal of Op- erations Management, Vol. 21, No. 3, pp. 329-351. Moledina, Amyaz A., Jay S. Coggins, Stephen Polasky, and Christopher Costello (2003) “Dynami Environmental Poli
y with Strategi
 Firms: Pri
es versus Quantities,” Journal of Environmental E
onomi
s and Management, Vol. 45, No. 2, pp. 356-376. Moore, Megan (2007) “Energy & Environmental Giving in the States,” resear
h report, The Na- tional Institute on Money in State Politi
s. Nakamura, Masao, Takuya Takahashi, and Ilan Vertinsky (2001) “Why Japanese Firms Choose to Certify: A Study of Managerial Responses to Environmental Issues,” Journal of Environmental E
onomi
s and Management, Vol. 42, No. 1, pp. 23-52. Newell, Ri
hard G. and William A. Pizer (2003) “Regulating Sto
k Externalities under Un
er- tainty,” Journal of Environmental E
onomi
s and Management, Vol. 45, No. 2, pp. 416-432. Nishitani, Kimitaka (2009) “An empiri
al study of the initial adoption of ISO 14001 in Japanese manufa
turing rms,” E
ologi
al E
onomi
s, Vol. 60, No. 3, pp. 669-679. OECD (1999) Voluntary Approa
hes for Enivronmental Poli
y: An Assessment, Paris, Fran
e. Potoski, Matthew and Aseem Prakash (2005a) “Covenants with weak swords: ISO 14001 and fa
ilities' environmental performan
e,” Journal of Poli
y Analysis and Management, Vol. 24, No. 4, pp. 745-769. (2005b) “Green 
lubs and voluntary governan
e: ISO 14001 rms' regulatory 
ompli- an
e,” Ameri
an Journal of Politi
al S
ien
e, Vol. 49, No. 2, pp. 235-248. 94 Quirion, Philippe (2004) “Regulating Sto
k Externalities under Un
ertainty,” Environmental and Resour
e E
onomi
s, Vol. 29, No. 3, pp. 337-359. Rennings, Klaus, Manuel Frondel, Jens Horba
h, and Till Requate (2004) “En- vironmental poli
y tools and rm-level management in Germany.” available at: http://www.oe
d.org/dataoe
d/26/17/31685533.pdf. Segerson, Kathleen and Thomas J. Mi
eli (1998) “Voluntary Environmental Agreements: Good or Bad News for Enviornmental Quality,” Journal of Environmental E
onomi
s and Management, Vol. 36, No. 2, pp. 109-130. (1999) “Voluntary Approa
hes to environmental prote
tion: The Role of Legislative threats,” in Carraro, Carlo and François Lévêque eds. Voluntary Approa
hes in Environmental Poli
y, Boston: Kluwer A
ademi
. Teikoku Data Bank (2003a) Company Yearbook East Japan: Teikoku Data Bank. (2003b) Company Yearbook West Japan: Teikoku Data Bank. (2004a) Company Yearbook East Japan: Teikoku Data Bank. (2004b) Company Yearbook West Japan: Teikoku Data Bank. (2005a) Company Yearbook East Japan: Teikoku Data Bank. (2005b) Company Yearbook West Japan: Teikoku Data Bank. US Environmental Prote
tion Agen
y (1999) 33/50 Program: The Final Re
ord., Washington DC: Of
e of Pollution Prevention and Toxi
s (7408) U.S. EPA. (2010) Risk-s
reening environmental indi
ators user's manual, Te
hni
al Appendix A, Listing of all toxi
ity weights for TRI 
hemi
als and 
hemi
al 
ategories. available at: http://www.epa.gov/opptintr/rsei/pubs/te
hni
al_appendixa_toxi
ity.pdf. Weitzman, Martin L. (1974) “Pri
es vs. Quantities,” Review of E
onomi
 Studies, Vol. 41, No. 4, pp. 477-491. 95 Wel
h, Eri
 and Akira Hibiki (2003) “An Institutional Framework for Analysis of Voluntary Poli
y: The Case of Voluntary Pollution Prevention Agreements in Kita Kyushu, Japan,” Journal of Environmental Planning and Management, Vol. 46, No. 4, pp. 523-554. Wel
h, Eri
 W., Yasuhumi Mori, and Midori Aoyagi-Usui (2002) “Voluntary adoption of ISO 14001 in Japan: me
hanisms, stages and effe
ts,” Business Strategy and the Environment, Vol. 11, No. 1, pp. 43-62. Ytterhus, Bjarne E. (2004) “Environmental poli
y tools and rm-level management in Norway.” available at: http://www.oe
d.org/dataoe
d/25/37/31684454.pdf. Yu, Zhihao (2005) “Environmental Prote
tion: A Theory of Dire
t and Indire
t Competition for Politi
al Inuen
e,” Review of E
onomi
 Studies, Vol. 72, No. 1, pp. 269-286. 96 Appendi
es Appendix A: Appendix for Chapter 2 A.1 A set of abatement rate and the number of parti
ipating rms that gener- ates the highest aggregate abatement if N V P is not an integer In this appendix, we show the set of abatement rates and the number of parti
ipating rms that generate the highest aggregate abatement when N V P is not an integer. Let N 0 P be an integer su
h that (N 0 P + e  1)a V = Na L where a L = 1  dē i N  
l dē i N 2 (1l ) ; a V = 1  dē i N  
l dē i N 2 (1l ) + 
l 2dē i N 3 (1l ) = a L + 
l 2dē i N 3 (1l ) ;and 0 < e < 1. This N 0 P is the greatest integer that satises (2.10) when the abatement rate satises (2.8) with equality. Let a 0V be the abatement rate su
h that N 0 P a 0V = Na L . We 
an then des
ribe the equilibrium abatement rate and the number of parti
ipat- ing rms. Proposition A.1. a 0V and N 0 P +1 are the equilibrium abatement rate and the number of parti
ipat- ing rms that generate the highest aggregate abatement if 2feN (1 e) 2 gfd(1l )ē i N 3  
N 2 (1l )N
lg (1 e) 2 
l > 0: (1) Otherwise, a V and N 0 P are the equilibrium abatement rate and the number of parti
ipating rms. Proof: From the argument given prior to Proposition 1, it is obvious that a V and N 0 P or a 0V and N 0 P + 1 are the equilibrium abatement rate and the number of parti
ipating rms. From simple 
omputation, we get a V N 0 P = a V (N 0 P + e1)+a V (1 e) = a L N+a V (1 e): (2) 97 and a 0V (N 0 P +1) = a 0V N 0 P +a 0V = a L N+a L N=N 0 P : (3) be
ause (N 0 P +e1)a V = Na L , and N 0 P a 0V = Na L . From Lemma 1, the regulator will 
hoose a 0V if a V N 0 P < a 0V (N 0 P +1). Therefore, the regulator will 
hoose a 0V if a 0V (N 0 P +1)a V N 0 P = a L N=N 0 P a V (1 e) = 1 N 0 P [Na L  (N a L a V +1 e)a V (1 e)℄ = 1 N 0 P [eNa L  (1 e) 2 a V ℄ = 1 N 0 P  [eN (1 e) 2 ℄(1  dē i N  
l d(1l )ē i N 2 ) (1 e) 2 
l 2d(1l )ē i N 3  = 2feN (1 e) 2 gfd(1l )ē i N 3  
N 2 (1l )N
lg (1 e) 2 
l 2N 0 P d(1l )ē i N 3 > 0: Therefore, a 0V (N 0 P +1)> a V N 0 P if (1) holds. QED A.2 Complete proof of the statement that there is equilibrium only when less than (N V P 1) rms parti
ipate in the VP and the mandatory poli
y is imple- mented (the se
ond part of proposition 2.2) The mandatory poli
y generates a (weakly) higher aggregate abatement if the abatement rate is a V and if the number of parti
ipating rms is less than (N V P 1) be
ause (N V P 1)a V ē i = Na L ē i . Therefore, the mandatory poli
y is implemented under su
h a 
ase. Suppose that N P (less than N V P 1) rms parti
ipate in the VP. As long as N P < N V P , the mandatory poli
y is implemented. Even if the VP is implemented, parti
ipating rms get the same payoff as they do when the mandatory poli
y is implemented. Therefore, if N P < N V P , then all parti
i- 98 pating rms do not have to 
hange their de
ision (from “parti
ipate” to “not parti
ipate”) In addi- tion, all “non-parti
ipating” rms do not have to 
hange their de
ision (from “not parti
ipate” to “parti
ipate”). QED A.3 Proof of Proposition 2.3, 2.5 and 2.6 We give proof of Propositions 2.3, 2.5, and 2.6 in the Appendix. We show the effe
ts of 
hanges in ea
h parameter on the abatement rate of the VP and the mandatory poli
y, the parti
ipation rate of the VP, and the aggregate abatement of the two programs. We havea L = 1  dē i N  
l dē i N 2 (1l ) ; a V = 1  dē i N  
l dē i N 2 (1l ) + 
l 2dē i N 3 (1l ) , N V P =N= a L a V +1= 1+1=N 
l 2dē i N 3 (1l )2N 2 (1l )
2N
l+
l , and N V P a V ē i = Na L ē i +a V ē i . Let A= 2dē i N 3 (1l ) and B= 
l 2dN 3 (1l ) . then, (ē i ) 0 ¶a V ¶ ē i =  dNē 2 i + 
l2dN 3 (1l ) A 2 (2N1)  dNē 2 i + 
l2dN 3 (1l ) A 2 2N= ¶a L ¶ ē i . ¶ (N V P =N) ¶ ē i =  a V  2 (a V ¶a L ¶ ē i  a L ¶a V ¶ ē i ) 0. ¶N V P a V ē i ¶ ē i = ¶Na L ē i ¶ ē i + ¶a V ¶ ē i ē i  ¶Na L ē i ¶ ē i . (N) 0  ¶a V ¶N =  dN 2 ē i + 
l2dN 2 ē i (1l ) A 2 (4N  3) =  dN 2 ē i + 
l6dN 2 ē i (1l ) A 2 (2N  1) 2
l A   dN 2 ē i + 
l6dN 2 ē i (1l ) A 2 2N = ¶a L ¶N . ¶N V P ¶N = a L a V +  a V  2 N[a V ¶a L ¶N a L ¶a V ¶N ℄ = 1  a V  2 [NB ¶a L ¶N + Na L ¶B ¶N  B(a L + B)℄  1 be
ause N ¶B ¶N = 3B and ¶a L ¶N = d dN 2 ē i + 4B. If ¶N V P ¶N  1, then ¶ (N V P =N) ¶N  0. Therefore, ¶ (N V P =N) ¶N  0. ¶N V P a V ē i ¶N = ¶Na L ē i ¶N + ¶a V ¶N ē i  ¶Na L ē i ¶N . (d) 0 ¶a V ¶d =  d 2 Nē i + 
l2ē i N 3 (1l ) A 2 (2N1)  d 2 Nē i + 
l2ē i N 3 (1l ) A 2 2N= ¶a L ¶d . ¶ (N V P =N) ¶d =  a V  2 (a V ¶a L ¶d  a L ¶a V ¶d ) 0. ¶N V P a V ē i ¶d = ¶Na L ē i ¶d + ¶a V ¶d ē i  ¶Na L ē i ¶d . (
) 0 ¶a V ¶ = 1 dNē i  l2dē i N 3 (1l ) A (2N1) 1 dNē 2 i  l2dē i N 3 (1l ) A 2N= ¶a L ¶ . ¶ (N V P =N) ¶ =  a V  2 (a V ¶a L ¶  a L ¶a V ¶ ) 0. ¶N V P a V ē i ¶ = ¶Na L ē i ¶ + ¶a V ¶ ē i  ¶Na L ē i ¶ . (l ) 0 ¶a V ¶l = 2
dē i N 3 A 2 (2N1) 2
dē i N 3 A 2 2N = ¶a L ¶l . ¶ (N V P =N) ¶l =  a V  2 (a V ¶a L ¶l a L ¶a V ¶l ) 0. ¶N V P a V ē i ¶l = ¶Na L ē i ¶l + ¶a V ¶l ē i  ¶Na L ē i ¶l . QED A.4 Proof of Proposition 2.4 When Nē i = N 0 ē 0 i ; N > N 0 ; 
 = 0 ; d = d 0 , and l = l 0 , a  = 1  dNē i = 1  dN 0 ē 0 i = a 0 . From a simple 
al
ulation, we get N V P =N = 1 
l 2dN 3 ē i (1l ) + 1=N and N V 0 P =N 0 = 1 
l 2dN 30 ē i (1l ) + 1=N 0 . 99 Let A = 
l 2dNē i (1l ) (= 
l 2dN 0 ē 0 i (1l ) ). Then, N V P =NN V 0 P =N 0 = A(1=N 2  1=N 02 )+ 1=N 1=N 0 = 1 N 2 N 02 (A(N 02 N 2 )NN 0 (N 0 N)) = N 0 N N 2 N 02 (A(N 0 +N)NN 0 ): As N 0 < N, N V P =NN V 0 P =N 0 > 0 if A(N 0 +N) < NN 0 . Be
ause we fo
us on a 
ase with a L0 = 1  dNē i  
l dN 2 ē i (1l ) = 1  dNē i  2A=N 0 > 0 as we mentioned in footnote 6 of Chapter 2, 2A= 
l dN 0 ē 0 i (1l ) < N 0 must hold. Therefore, A(N 0 +N)< 1 2 N 0 (N 0 +N)< NN 0 and N V P =NN V 0 P =N 0 > 0 be
ause N > N 0 . Be
ause a V = 1  dNē i  
l dN 2 ē i (1l ) + 
l 2dN 3 ē i (1l ) = 1  dNē i 2A=N+A=N 2 and a V 0 = 1  dN 0 ē 0 i  
l dN 02 ē 0 i (1l ) + 
l 2dN 03 ē 0 i (1l ) = 1  dN ¯ 0 e i  2A=N 0 + A=N 02 , a V  a V 0 = 2A=N +A=N 2 + 2A=N 0  A=N 02 = 2A=N 2 N 02 (N 02 N 2  2N 0 N(N 0 N)) = 2A(N 0 N)=N 2 N 02 (N 0 +N 2N 0 N). Be
ause N > N 0  2, (N 0 N)< 0 and (N 0 +N2N 0 N)< 0: Thus, a V a V 0 > 0. QED A.5 Case without free ride in lobbying If we assume that all rms 
ooperate in lobbying and that no rms free-ride, then the lobby group minimizes C(W) = å N i=1 
a(W)ē i +W subje
t to ¶L ¶W = 0 and L(a;W) = L(a  ;0). From the F.O.C and L(a;W) = L(a  ;0), we obtain a L = 1  dNē i (1l ) andW=  2 l 2d(1l ) . If we assume the politi
al 
ontribution is paid equally by all rms, then a V = 1 
(2l ) 2dNē i (1l ) and N V P =N = a L a V +1=N. Under a 
ase with free-riding in lobbying, we obtain the following result, whi
h is 
ompletely different from that without free-riding. Proposition A.2. If Nē i = N 0 ē 0 i ; N > N 0 ; 
 = 0 ; d = d 0 , and l = l 0 , then a V = a V 0 , N V P =N < N V 0 P =N 0 , and N V P a V ē i < N V 0 P a V 0 ē 0 i . Proof: Be
ause Nē i = N 0 ē 0 i , a V = 1 
(2l ) 2dNē i (1l ) = 1 
(2l ) 2dN 0 ē 0 i (1l ) = a V 0 . N V P =N = a L a V + 1=N < a L0 a V 0 + 1=N 0 = N V 0 P =N 0 as a L = a L0 ,a V = a V 0 and N < N 0 . N V P a V ē i = Na L ē i +a V ē i < N 0 a L0 ē 0 i + a V 0 ē 0 i = N V 0 P a V 0 ē 0 i as ē i < ē 0 i , a L = a L0 ,a V = a V 0 and Nē i = N 0 ē 0 i . QED Proposition A.3. (1) ¶ (N V P =N) ¶ ē i  0, ¶N V P a V ē i ¶ ē i  0, ¶a V ¶ ē i  0, (2) ¶ (N V P =N) ¶N  0, ¶a V ¶N  0, ¶N V P a V ē i ¶N  0, (3) ¶ (N V P =N) ¶d  0, ¶a V ¶d  0, ¶N V P a V ē i ¶d  0, (4) ¶ (N V P =N) ¶  0, ¶a V ¶  0, ¶N V P a V ē i ¶  0, and (5) ¶ (N V P =N) ¶l  0, ¶a V ¶l  0, ¶N V P a V ē i ¶l  0. 100 Proof: We have a L = 1  dNē i (1l ) , a V = 1 
(2l ) 2dN(1l )ē i , N V P =N = a L a V + 1=N and N V P = [1 
l 2dN(1l )ē i 
(2l ) ℄N+1. From a simple 
al
ulation, we nd DSC=a V ē i ( 1 2 da V ē i  
l N(1l ) )=a V ē i [ 1 2 dē i   2N  
l 2N 2 (1l ) + 
l 4N 3 (1l )  
l N(1l ) ℄> 0: Hen
e, (ē i ) 0 ¶a V ¶ ē i = 
(2l ) 2dN(1l )ē 2 i  2 2dN(1l )ē 2 i = ¶a L ¶ ē i . ¶ (N V P =N) ¶ ē i =  a V  2 (a V ¶a L ¶ ē i a L ¶a V ¶ ē i ) 0. ¶N V P a V ē i ¶ ē i = ¶Na L ē i ¶ ē i + ¶a V ¶ ē i ē i  ¶Na L ē i ¶ ē i . (N) 0 ¶a V ¶N = 
(2l ) 2dN 2 (1l )ē i  2 2dN 2 (1l )ē i = ¶a L ¶N . ¶N V P ¶N = 1+  2 l (2l ) [2dN(1l )ē i 
(2l )℄ 2  1. If ¶N V P ¶N  1, then ¶ (N V P =N) ¶N  0. Therefore, ¶ (N V P =N) ¶N  0. ¶N V P a V ē i ¶N = ¶Na L ē i ¶N + ¶a V ¶N ē i  ¶Na L ē i ¶N . (d) 0 ¶a V ¶d = 
(2l ) 2d 2 N(1l )ē i  2 2d 2 N(1l )ē i = ¶a L ¶d . ¶ (N V P =N) ¶d =  a V  2 (a V ¶a L ¶d a L ¶a V ¶d ) 0. ¶N V P a V ē i ¶d = ¶Na L ē i ¶d + ¶a V ¶d ē i  ¶Na L ē i ¶d . (
) 0 ¶a V ¶ = 2+l 2dN 2 (1l )ē i  2 2dN 2 (1l )ē i = ¶a L ¶ . ¶ (N V P =N) ¶ =  a V  2 (a V ¶a L ¶ a L ¶a V ¶ ) 0. ¶N V P a V ē i ¶ = ¶Na L ē i ¶ + ¶a V ¶ ē i  ¶Na L ē i ¶ . (l ) 0 ¶a V ¶l =  2dN(1l )ē i  2 2dN 2 (1l )ē i = ¶a L ¶l . ¶ (N V P =N) ¶l =  a V  2 (a V ¶a L ¶l a L ¶a V ¶l ) 0. ¶N V P a V ē i ¶l = ¶Na L ē i ¶l + ¶a V ¶l ē i  ¶Na L ē i ¶l . QED 101 Appendix B: Appendix for Chapter 3: Estimation results of WM IV and V models with m= 3;5; and 10 We show the estimation results of WM IV and V models with m=3, 5, and 10. We show the results of ISO 14001 adoption by water emitting fa
ilities (Table B.1), water emission redu
tion (Table B.2), ISO 14001 adoption by air emitting fa
ilities (Table B.3), and air emission redu
tion (Table B.4). 102 Table B.1: Estimation results of ISO 14001 adoption by water emitting fa
ilities with different 'm's Dependent variable:ISO 14001 adoption WM IV WM V m=3 m=5 m=10 m=3 m=5 m=10 Workers/10 2 0.119  0.096  0.094  0.134  0.128  0.109  (0.011) (0.009) (0.015) (0.013) (0.012) (0.014) Air emission intensity/10 6 0.056 0.031 0.075 0.018 0.034 0.029 (0.270) (0.265) (0.271) (0.270) (0.263) (0.254) Water emission intensity/10 6 -0.254  -0.145  -0.137  -0.208  -0.144  -0.147  (0.096) (0.073) (0.076) (0.123) (0.080) (0.084) Prot/revenue/10 3 1.11  0.924 0.912 0.958 0.949 0.923 (0.656) (0.648) (0.652) (0.640) (0.625) (0.665) Population density/10 3 -5.62 -5.08 -3.81 -4.83 -4.73 -4.36 (3.47) (3.51) (3.57) (3.52) (3.41) (3.56) In
ome per 
apita/10 3 0.293 0.238 0.223 0.268 0.318 0.242 (0.375) (0.349) (0.362) (0.369) (0.372) (0.376) Support -0.038 -0.053 -0.032 -0.019 -0.022 -0.016 (0.136) (0.134) (0.140) (0.137) (0.133) (0.144) Grievan
es against pollution -0.500 -0.346 -0.612 -0.462 -0.558 -0.481 (0.455) (0.458) (0.463) (0.464) (0.427) (0.468) Year dummy (2001) -0.227  -0.211  -0.200  -0.269  -0.258  -0.225  (0.075) (0.074) (0.073) (0.082) (0.085) (0.078) Constant -19.8 -13.6 -29.6  -32.7  -34.2 -7.44 (12.5) (10.1) (10.5) (16.3) (27.1) (4.78) n 0.333  0.357  0.334  0.048 0.043 0.166  (0.046) (0.038) (0.067) (0.032) (0.054) (0.054) Likelihood -1237.74 -1411.22 -1121.92 -1101.20 -1465.35 -1099.74 Observation 1326 1326 1326 1326 1326 1326 All of models in this table are SAR probit models.  ,  , and  imply that the 
oef
ient is signi
antly different from zero at the 1%, 5% and 1% levels, respe
tively. 103 Table B.2: Estimation results of water emissions redu
tions with different 'm's Dependent variable: Water emissions redu
tions (E W t+1 =E W t ) WM IV WM V m=3 m=5 m=10 m=3 m=5 m=10 Workers -0.367 -0.368 -0.366 -0.367 -0.367 -0.367 (1.73) (1.73) (1.73) (1.73) (1.73) (1.74) Prot/Revenue -0.352 -0.352 -0.350 -0.351 -0.351 -0.355 (4.51) (4.51) (4.52) (4.52) (4.52) (4.52) ISO 176 174 172 172 172 173 (847) (849) (849) (849) (849) (849) Population density -1171 -1167 -1167 -1167 -1167 -1169 (1742) (1745) (1746) (1746) (1746) (1746) In
ome per 
apita 2.22 2.21 2.21 2.20 2.20 2.21 (12.1) (12.1) (12.1) (12.1) (12.1) (12.1) Grievan
es against pollution -1.92 -1.89 -1.90 -1.90 -1.90 -1.91 (10.1) (10.2) (10.2) (10.2) (10.2) (10.2) Constant -134 -134 -134 -135 -135 -135 (487) (488) (488) (488) (488) (489) r 0.006 0.002 0.001 0.0003 0.0005 -0.003 (0.000) (0.000) (0.000) (0.0000) (0.0000) (0.000) Log-likelihood -6647.038 -6647.041 -6647.041 -6647.041 -6647.041 -6647.041 Observation 663 663 663 663 663 663 All of models in this table are xed effe
ts SAR models.  ,  , and  imply that the 
oef
ient is signi
antly different from zero at the 1%, 5% and 1% levels, respe
tively. 104 Table B.3: Estimation results of ISO 14001 adoption by air emitting fa
ilities with different 'm's Dependent variable: ISO 14001 adoption WM IV WM V m=3 m=5 m=10 m=3 m=5 m=10 Workers/10 2 0.129  0.107  0.097  0.122  0.107  0.085  (0.006) (0.005) (0.016) (0.006) (0.005) (0.006) Air emission intensity/10 6 0.167 0.16 0.163 0.169 0.154 0.165 (0.129) (0.263) (0.130) (0.127) (0.130) (0.125) Water emission intensity/10 6 -0.127 -0.138 -0.139 -0.130 -0.121 -0.118 (0.105) (0.102) (0.107) (0.123) (0.11) (0.102) Prot/revenue/10 3 0.003 -0.053 -0.050 0.097 0.072 0.061 (0.248) (0.220) (0.220) (0.236) (0.229) (0.233) Population density/10 3 -1.78 -2.07  -1.95 -1.65 -1.40 -1.20 (1.149) (1.184) (1.24) (1.18) (1.11) (1.17) In
ome per 
apita/10 3 -0.150 -0.144 -0.216 -0.120 -0.113 -0.129 (0.121) (0.122) (0.123) (0.117) (0.118) (0.121) Support 0.070 0.078 -0.054 0.068 0.065 0.059 (0.050) (0.049) (0.046) (0.049) (0.046) (0.048) Grievan
es against pollution -0.289  -0.281 -0.235 -0.322 -0.345  -0.355  (0.171) (0.18) (0.182) (0.170) (0.163) (0.171) Year dummy (2001) -0.178  -0.144  -0.130  -0.181  -0.161 -0.131  (0.030) (0.03) (0.035) (0.032) (0.03) (0.028) Constant -3.38  -3.58  -54.5  -4.03  -4.198  -3.02  (1.78) (1.99) (21.4) (2.08) (2.03) (1.26) n 0.270  0.419  0.430  0.199  0.287  0.417  (0.000) (0.001) (0.090) (0.020) (0.019) (0.025) Likelihood -6658.74 -6614.50 -6352.29 -6523.54 -6461.20 -6337.96 Observation 7158 7158 7158 7158 7158 7158 All of models in this table are SAR probit models.  ,  , and  imply that the 
oef
ient is signi
antly different from zero at the 1%, 5% and 1% levels, respe
tively. 105 Table B.4: Estimation results of air emissions redu
tions with different 'm's Dependent variable: Air emissiions redu
tions (E A t+1 =E A t ) WM IV WM V m=3 m=5 m=10 m=3 m=5 m=10 Workers 0.0242 0.0249 0.0248 0.0248 0.0248 0.0256 (0.0514) (0.0515) (0.0523) (0.0528) (0.0528) (0.0526) Prot/Revenue 0.004 0.0069 0.0062 0.0163 0.0162 0.0189 (0.098) (0.098) (0.100) (0.100) (0.100) (0.100) ISO -14.6 -13.8 -12.9 -13.1 -13.1 -13.3 (20.6) (20.1) (20.4) (20.6) (20.6) (20.5) Population density 6.04 5.21 5.72 6.41 6.43 5.97 (26.0) (26.1) (26.4) (26.7) (26.7) (26.6) In
ome per 
apita -0.1705 -0.169 -0.1672 -0.178 -0.178 -0.1761 (0.227) (0.227) (0.231) (0.233) (0.233) (0.232) Grievan
es against pollution 0.133 0.131 0.13 0.144 0.144 0.142 (0.196) (0.197) (0.200) (0.202) (0.202) (0.201) Constant 6.44 6.06 6.03 6.72 6.73 6.36 (10.5) (10.5) (10.7) (10.8) (10.8) (10.7) r 0.075  0.110  0.090  0.0001  -0.001  0.039  (0.000) (0.000) (0.000) (0.0000) (0.000) (0.000) Likelihood -25422.42 -25419.32 -25423.81 -25426.65 -25426.65 -25426.18 Observation 3579 3579 3579 3579 3579 3579 All of models in this table are xed effe
ts SAR models.  ,  , and  imply that the 
oef
ient is signi
antly different from zero at the 1%, 5% and 1% levels, respe
tively. 106 Appendix C: Appendix for Chapter 4 C.1 Cal
ulations of t HIC and t LIC Suppose 1 1 2 F(ē H )<C H (e H (
ē L ))+ 
ē L e H (
ē L )+W L (
ē L ): (4) t HIC L must satisfy 1 2 F(ē H ) =C H (e H (t HIC L ))+ t L e H (t HIC L )+W L (t HIC L ) (5) where 1 2 F(ē i ) = t C i e C i (t C i )+C i (e C i (t C i ))+W C i (t C i ). By rearranging (5), ((a1)
+a)t 2 2
(a ē L  
ē H )t+ 2  a 1+ ē 2 L F(ē H )  = 0 (6) Let A T = (a1)
+a;B T = 
(a ē L  
ē H ), andC T = 2  a 1+ ē 2 L F(ē H )  . Then, t HIC L = B T + p B 2 T A T C T A T : Next, suppose (4) does not hold. By rearranging (5), t 2 2
ē H t+ 
[F(ē H )2W L (
ē L )℄ = 0: (7) LetC 0 = 
[F(ē H )W L (
ē L )℄. Then, t HIC L = 
ē H  p (
ē H ) 2 C 0 be
ause 0 t HIC L  
ē H and C 0 > 0. t HIC L exists as long as (
ē H ) 2 C 0 > 0, i.e., 1 This is the 
ondition for t HIC L < 
ē L . 107 12 F(ē H )<C H (0)+W L (
ē L ): (8) Next, t LIC L must satisfy C L (e L (t C H ))+ t C H e L (t C H )+W C H (t C H ) =C L (e L (t LIC L ))+ t LIC L e L (t LIC L )+W L (t LIC L ) (9) ifC L (e L (t C H ))+ t C H e L (t C H )+W C H (t C H )<C L (0)+W L (
ē L ). Otherwise, t LIC L = 
ē H . C.2 Contribution under pooling equilibria when (1+ 
)ē L < E[ē℄ and when t < 
ē L and (1+ 
)ē L  E[ē℄ In this 
ase 2 , the so
ially optimal pooling tax rate is the same as the so
ially optimal tax rate for the high-
ost industry. Hen
e, the 
ontribution should be W(t) = 8 > < > : a 2 h t  2 + p L (ē L  t  ) 2 +(1 p L )(ē H  t  ) 2 fp L 
ē 2 L +(1 p L )
=(1+ 
)ē 2 H g i i f t < 
ē L a 2 h p L 
ē 2 L +(1 p L )f t  2 +(ē H  t  ) 2 gfp L 
ē 2 L +(1 p L )
=(1+ 
)ē 2 H g i otherwise: (10) If (a1)
+a  0, then the best pooling equilibrium tax rate is t P = 0. Otherwise, the best pooling equilibrium tax rate is t P = t  H if t P 2 > t  H (otherwise, t P = t P 2 ). If t > 
ē L and (1+ 
)ē L  E[ē℄, then the 
ontribution should be W(t) = a 2  p L 
ē 2 L +(1 p L )f t  2 +(ē H  t  ) 2 gfp L 
=(1+ 
)ē 2 L +(1 p L )
=(1+ 
)ē 2 H g  : (11) 2 This 
ondition is the same as 
ē L < (E[ē℄a
ē L ) (a1)
+a 108 C.3 Contribution under pooling equilibria when (1+ 
)ē L < 
E[ē℄ and when q> ē L and (1+ 
)ē L  
E[ē℄ If (1+ 
)ē L < 
E[ē℄, the so
ially optimal pooling quota is the same as the so
ially optimal quota for the high-
ost industry. Hen
e, the 
ontribution should be W(q) = a(1 p L ) 2(1+ 
)  (1+ 
)q 
ē 2 H  2 : (12) And, the best pooling equilibrium quota is q P 2 . If ē L < q and (1+ 
)ē L  
E[ē℄, then the 
ontribution should be W(q) = a 2  p L ē 2 L +(1 p L )(
(ē H q) 2 +q 2 )f
E[ē 2 ℄ (
E[ē℄) 2 1+ g  : (13) C.4 Proof of Proposition 4.1 TC q =C i (q C i )+W i (q C i ) = a 2(1+ 
)A ē 2 i : (14) TC t =C i (e i (t C i ))+ t C i e i (t C i )+W i (t C i ) = 8 > < > : a 2(1+
) ē 2 i i f a  a+2a

 2 2(1+
)A T ē 2 i otherwise (15) where A = 
+a(1+ 
) and A T = 
+a(1+ 
). If a  
, TC q < TC t be
ause 
=A < 1. Next, 
onsider a > 
. TC t TC q = a+2a
 
 2 2(1+ 
)A T ē 2 i  a 2(1+ 
)A ē 2 i : = (a+2a
 
 2 )Aa
A T 2(1+ 
)AA T ē 2 i = (a 
)(1+ 
)A+a 2 +(1+ 
)a 2 
+a 2 a 2  2 a 2  2(1+ 
)AA T ē 2 i = (a 
)(1+ 
)A+2a 2 2(1+ 
)AA T ē 2 i > 0: 109 Hen
e, TC t > TC q . QED C.5 Proof of Proposition 4.2 proof of( Suppose that a separating equilibrium exists but t C H > ˆ t. Then, be
ause there is a separating equi- librium, there is ˜ t su
h that C H (e H (t C H ))+ t C H e H (t C H )+W H (t C H )C H (e H ( ˜ t))+ ˜ te H ( ˜ t)+W L ( ˜ t) (16) C L (e L ( ˜ t))+ ˜ te L ( ˜ t)+W L ( ˜ t)C L (e L (t C H ))+ t C H e L (t C H )+W H (t C H ) (17) Be
ause t C H > ˆ t,W L ( ˆ t) =W H ( ˆ t)>W H (t C H ), and therefore,W L (t)>W H (t C H ) for all t  ˆ t. if ˜ t > t C H > ˆ t, then (17) does not hold be
ause W L ( ˜ t) > W L ( ˆ t) > W H (t C H ) and C i and te i (t) are nonde
reasing on T . Thus, t C H > ˜ t must hold. However, if 
ē L  t C H , we get t C H  ˜ t by adding (17) to (16) be
ause C L (e L (t)) = C H (e H (t)) and te L (t) te H (t) = t ¯ (e H  ¯ e L ) when t < 
ē L . This is a 
ontradi
tion. If ˜ t  
ē L < t C H , then by adding (17) to (16), we get (t C H ) 2 2 + t C H (ē H  t C H  ) 1 2 
ē 2 L  ˜ tDē t C H Dē+ 1 2 [ē L (t C H  
ē L ) t C H  (t C H  
ē L )℄ ˜ tDē (t C H  ˜ t)
Dē 1 2 (t C H  
ē L ) 2 : Be
ause 
ē H > t C H , 
Dē > t C H  
ē L . Hen
e, t C H  ˜ t < (t C H  
ē L )=2. By rearranging, ˜ t > (t C H + 
ē L )=2> 
ē L . This is a 
ontradi
tion. If 
ē L  ˜ t < t C H , (16) implies W L ( ˜ t) = W L (
ē L )  W H (t C H ). W L ( ˜ t)  W H (t C H ) and (17) imply that t C H  ˜ t. This is a 
ontradi
tion. Therefore, t C H  ˆ t if there exists a separating equilibrium. Proof of) 110 Be
ause t C H  ˆ t, there exists  t, t C H   t  ˆ t, su
h thatW H (t C H )W L (  t) andC H (e C H )+t C H e C H +W H (t C H ) = C H ( e H )+  t e H +W L (  t). Case: t C H  ˆ t  
ē L C L (e L (  t))+  te L (  t)+W L (  t) =C H (e H (  t))+  te H (  t)+W L (  t)  tDē =C L (e L (t C H ))+ t C H e L (t C H )+W H (t C H )  tDē =C L (e C L )+ t C H e C L +W H (t C H )+(t C H   t)Dē C L (e L (t C H ))+ t C H e L (t C H )+W H (t C H ) Hen
e, ICs for both types hold in this 
ase. Case:  t  t C H  
ē L C L (e L (  t))+  te L (  t)+W L (  t) =C H (e H (  t))+  te H (  t)+W L (  t)  te H (  t)C H (e H (  t))+C L (e L (t C H )) =C H (e H (t C H ))+ t C H e H (t C H )+W H (t C H )  te H (  t)C H (e H (  t))+C L (e L (t C H )) =C L (e L (t C H ))+ t C H e L (t C H )+W H (t C H )  [C H (e H (  t))+  te H (  t)C H (e H (t C H )) t C H e H (t C H )℄ C L (e L (t C H ))+ t C H e L (t C H )+W H (t C H ) Case:  t  
ē L  t C H C L (e L (  t))+  te L (  t)+W L (  t) =C H (e H (  t))+  te H (  t)+W L (  t)  te H (  t) 1 2 (  t 2  (
ē L ) 2 ) =C H (e H (t C H ))+ t C H e H (t C H )+W H (t C H )  te H (  t) 1 2 [  t 2  (
ē L ) 2 ℄ =C L (e L (t C H ))+ t C H e L (t C H )+W H (t C H ) 1 2 [  t 2  (
ē L ) 2 ℄ [  te H (  t) t C H Dē℄ If 1 2 [  t 2  (
ē L ) 2 ℄+ [  te H (  t) t C H Dē℄ 0, C L (e L (  t))+  te L (  t)+W L (  t)C L (e L (t C H ))+ t C H e L (t C H )+W H (t C H ): (18) 111 We show this. 1 2 [  t 2  (
ē L ) 2 ℄+ [  te H (  t) t C H Dē℄ = (  t t C H )Dē 1 2 
ē 2 L  1 2  t 2 +  tē L = (  t t C H )Dē+ 1 2 (  t 
ē L )ē L  1 2  t(  t 
ē L ) = (  t t C H )Dē 1 2 (  t 
ē L ) 2  0: This is be
ause  t t C H   t
ē L , and 2
Dē  t+
ē L = 2
ē H   t
ē L  0. Hen
e, ICs for both types hold and there is a separating equilibrium. QED C.6 Proof of Proposition 4.4 By rearranging (4.13) and (4.14), f (q) = [
+(1+ 
)a℄q 2 2(
ē L +a
ē H )q+ 
ē 2 L + a 2 1+ ē 2 H G(ē L ) 0 (19) g(q) = (1+ 
)aq 2 2a
ē H q+ a 2 1+ ē 2 H G(ē L ) 0 (20) h(q) = [
+(1+ 
)a℄q 2 2(
ē H +a
ē H )q+ 
ē 2 H + a 2 1+ ē 2 H  (G(ē L )+H) 0 (21) where G(ē i ) = C L (e i (q C i ))+W C i (q C i ) and H = C H (e H (q C L ))C L (e L (q C L )). We get (19) and (20) by rearranging (4.13) and get (21) from (4.14). When q C H > ē L , the IC 
ondition for the low-
ost industry might be (20). Let A = 
+(1+ 
)a;A 00 = (1+ 
)a , B 0 = 
ē L +a
ē H , B = 
ē H +a
ē H , C 0 = 
ē 2 L + a 2 1+ ē 2 H G(ē L ), C 00 = a 2 1+ ē 2 H G(ē L ), and C = 
ē 2 H + a 2 1+ ē 2 H  (G(ē L )+H). Then, a separaing equilibrium q must satisfy q q 0 = B 0 + p B 0 2 AC 0 A < ē L (22) q q 00 = a
ē H + q (a
ē H ) 2 A 00 C 00 A 00 =  1+ ē H + p 
=A 1+ ē L < ē H (23) q q 2 = B+ p B 2 AC A : (24) 112 where f (q 0 ) = 0, g(q 00 ) = 0, and h(q 2 ) = 0. There exists q satisfying the IC 
ondition for the high-
ost industry be
ause B 2 AC = A(G+H)+(
(1+a)ē H ) 2  (
+(1+ 
)a)
ē 2 H  (
+(1+ 
)a)a 2 (1+ 
) ē 2 H = A(G+H) a 1+ ē 2 H (25) = (
+a
)
Dē 2 + a 2 1+ (Dē 2 +2Dēē L ) = (
+a
)
Dē 2 + a 2 1+ (ē 2 H  ē 2 L )> 0 (26) where G = a (1+
)A ē 2 L , H = 
Dē 2 + 2 a A Dēē L , Dē = ē H  ē L , and Dē 2 = (ē H  ē L ) 2 . Suppose that q satisfying the IC 
ondition for the high-
ost industry is given by (20). Then, the smallest one is q 00 , g(q 00 ) = 0, and g(ē L ) 0. h(q 00 )< 0 be
ause g(ē L )h(ē L ) =
ē 2 L +2
ē H ē L  
ē 2 H +H =
Dē 2 + 
Dē 2 +2 a A Dēē L > 0 (27) ¶g ¶q = 2(1+ 
)aq2a
ē H (28) ¶h ¶q = 2
q+2(1+ 
)aq2
ē H 2a
ē H = ¶g ¶q +2
q2
ē H < ¶g ¶q if q< ē H : (29) Hen
e, q 00 < q 2 . Next, 
onsider the 
ase when the IC 
ondition for the high-
ost industry is given by (19). If B 02  AC 0 < 0, then any q satisfying (24) is a separating equilibrium. Next, we 
onsider when B 02 AC 0  0. q 0  q 2 implies that there exists a separating equilibrium. Now, we show it. B 2 B 0 2 = 2 Dē(ē H + ē L +2a ē H ) (30) ACAC 0 = A(
Dē(ē H + ē L ) 
Dē(ē H + ē L 2q C L )) (31) = 2(A
Dēq C L ) = 2 2 (1+a)Dēē L (32) B 2 B 0 2  (ACAC 0 ) = 2a 2 Dē(ē H  ē L )+ 2 Dē(ē H  ē L ) = (2a+1)(
Dē) 2 > 0: (33) 113 Hen
e, q 0  q 2 . QED 114

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0072234/manifest

Comment

Related Items