Dynamic Games and Contracts as Agreements

Bård Harstad

Minicourse Lecture Notes 2 — 2017

http://www.sv.uio.no/econ/personer/vit/bardh/dokumenter/uphd2.pdf

Abstract
This lecture note studies stochastic games and contract theory, and illustrates
how to combine the two. We will discuss:
(1) Dynamic and stochastic games
(2) Incomplete contracts and hold-up problems
(3) Optimal contracts and duration
(4) Renegotiation and renegotiation design
(5) Complete contracts, spillovers, and R&D subsidies

1. Introduction

These notes study dynamic games where both pollution stocks and technology stocks
cumulate over time. Furthermore, and more importantly, the notes allow for so-called
"legally binding agreements," where we assume that countries may be able to (in part)
commit to certain emission levels.
If countries can commit to any future action (emission as well as technology), then
the …rst best can easily be achieved. In reality, however, commitment is only imperfect:
Countries may not be able to commit to the very long term, and some actions may not
be possible to measure or observe, or commit to, at all.
For example, the importance of developing new and green technology has been recog-
nized in the treaty texts, but the "technology needs must be nationally determined, based
on national circumstances and priorities."1 Thus, both past and future agreements are
likely to prescribe emission levels rather than investments in technology. To some extent,
commitments on emissions will motivate countries to invest in new technology. However,
there are large externalities or technological spillovers associated with such investments.
When surveying the literature, IPCC (2014, Ch. 13:50) concludes that "the evidence
indicates a systematic [positive] impact of IP protection on technology transfer." It may
thus be comforting that the Agreement on Trade Related Aspects of Intellectual Property
Rights (TRIPS) of 1994 generally commits all countries to create and enforce standard
intellectual property rights. TRIPS, however, allows for exceptions to the exclusive rights
of patents for public policy reasons. It provides for the possibility of compulsory licensing
and royalty-free compulsory licensing has indeed been advocated for as a way to encourage
technology transfers (IPCC, 2014, chap. 13:50; 15:47). Also the 2015 Paris agreement fo-
cused on emissions, while investments in new technology was not something anyone tried
to specify. Furthermore, the commitments under the Paris agreement is only to 2025 (for
some countries) or 2030 (for others).
This raises several important questions. How valuable is such an agreement with a
relatively short duration? What is the optimal term of this agreement? How should the
term, and the design more generally, be in‡uenced by actual intellectual property rights
and the requirements for licensing?
These lecture notes present a dynamic framework in which countries both pollute
and invest in technology during every period. The pollution as well as the technology
stocks depreciate but accumulate over time. The technology, which could involve either
renewable energy sources or abatement technology, reduces the need to pollute. While
the model has a large number of subgame perfect equilibria, I focus on the symmetric
Markov perfect equilibrium (MPE) since it is simple, robust, and unique. If countries
cannot commit to any future action, the game is a dynamic version of the common-pool
problem. If countries can contract on every future action, they easily implement the …rst-
best outcome. The more realistic and interesting situation arises when countries contract
on emission but not investment levels.
1
The quote is from §114 in the Cancun Agreement (UNFCCC, 2011), con…rmed by the Durban
Platform (UNFCCC, 2012).

2
 While these notes abstract from more general functional forms, heterogeneity among
the countries, and renegotiation, they intend to make a number of important policy
lessons:
First, climate agreements can reduce welfare relative to business as usual. When fu-
ture negotiations are anticipated, countries may fear being held up by the others when
negotiating emission quotas.2 This hold-up problem reduces the incentive to invest, and
every country may be worse o¤ with short-term agreements than in the business-as-usual
scenario in which no negotiations or agreements ever occur. Speci…cally, climate agree-
ments are likely to be harmful if intellectual property rights are weak, the commitment
period is short, and the number of countries large.
Following the pessimistic result above, the optimal climate treaty is characterized.
If quotas are negotiated before a country invests, it cannot be held up by the other
countries–at least not before the agreement expires. Thus, countries invest more when
the agreement is long-term. Nevertheless, countries underinvest compared to the opti-
mum if technological spillovers are positive or intellectual property rights are weak. To
compensate for this and to encourage further investments, the best (and equilibrium)
treaty is tougher, in that it stipulates lower emissions, relative to what is optimal ex post,
once the investments are sunk. The weaker the intellectual property rights, the tougher
the optimal (and equilibrium) treaty.
Finally, the optimal length of the agreement is characterized. On the one hand, a
longer time horizon is required to minimize the hold-up problem and to maximize the
incentive to invest in technology. On the other hand, the future marginal cost of pollution
is uncertain and stochastic in the model, and it is hard to predict the ideal quotas in the
far future. The optimal length trades o¤ these concerns. If intellectual property rights
are strengthened, for example, the optimal length decreases.
The optimal climate treaty is a function of trade policies, but the reverse is also true: if
the climate treaty is relatively short-term, it is more important to strengthen intellectual
property rights, reduce tari¤s, or subsidize technological trade. Negotiating such trade
2
The New York Times reports that "Leaders of countries that want concessions say that nations like
Denmark have a built-in advantage because they already depend more heavily on renewable energy"
(October 17, 2008: A4).

3
policies is thus a strategic substitute for a tough climate agreement.
By analyzing environmental agreements as incomplete contracts in a dynamic game,
I contribute to three strands of literature.
The model is presented in the next section. Section 3 presents the (complete contract-
ing) …rst-best outcome as well as the (noncooperative) business-as-usual scenario. The
fact that short-term agreements can be harmful is shown in Section 4, while Section 5
characterizes the optimal agreement. Section 6 discusses trade policies and shows that
the main results hold whether or not side transfers are available or the emission permits
are tradable. The …nal section concludes and the proofs follow in the appendix.
The model is presented in the next section. When solving the model in Section 3, we
gradually increase the possibilities for negotiations and contracts: (i) no cooperation, (ii)
short-term agreements, (iii) long-term agreements, and (iv) long-term agreements with
renegotiation. We start out by assuming that investments are noncontractible, technolo-
gies nontradable, quotas nontradable, …rms nonexistent, and countries homogeneous. We
will discuss how these assumptions can be relaxed in Sections 4 and 5. Further readings
are discussed in Section 6. Most proofs are in the the Appendix s.

2. Stochastic Games

2.1. Payo¤-Relevant Stocks and Uncertainty

This section presents a game where n players over time contribute to a public good and
invest in technology. The purpose of the technology is to reduce the cost of providing
public goods in the future. While the model is general, let climate change …x ideas. I will
thus refer to the players as countries, the public good (or its negative: the public bad) as
the stock of greenhouse gases, and to contributions as emissions. For the most part, we
will abstract from heterogeneities within countries and oil exhaustability. The strategic
e¤ects studied below would survive if these complications were added to the model.
While a repeated game looks identical at the start of every period, a dynamic game,
and a stochastic game, can change from period to period in accordance with variables
described the state or the payo¤-relevant stocks.

4
 The public bad is represented by the stock G of greenhouse gases in excess of its natural
level. Since the natural level is thus G = 0, G tends to approach zero over time (were it
not for emissions), and 1 qG 2 [0; 1] measures the fraction of G that "depreciates" every
period. G may increase, nevertheless, depending on the emission level gi from country
i 2 f1; :::; ng:
X
G = qG G + + gi : (2.1)
i

G represents the stock of greenhouse gases left from the previous period; subscripts for
periods are thus skipped. The shock , arbitrarily distributed with mean 0 and variance
2
, captures Nature’s stochastic impact on G. Note that, although the shock t is i.i.d.
across periods, it has a long-lasting impact through its e¤ect on G (or on ). The additive
form in equation (2.1) is standard in the literature.3 We will abstract from country-speci…c
uncertainty.
The other type of stock is technology. The technology stock in country i is measured
by Ri , and it depreciates over time at the rate 1 qR 2 [0; 1]. If country i invests
ri units in technology, Ri increases directly by dri units and, allowing for technological
spillovers, Rj may increase by eri , 8j 6= i. Such spillovers are empirically important (see
Coe and Helpman, 1995, or Golombek and Hoel, 2004, for several references in the …eld
of environmental economics.)
X
Ri = qR Ri; + dri + erj : (2.2)
j6=i
P
Assuming the spillover is imperfect, d > e. The total e¤ect on R i Ri is de…ned
by the primitive constant D d + e (n 1). The simple externality e may represent
traditional technological spillovers, di¤usion, imitation, licensing, or trade. In particular,
if traditional measures of intellectual property rights (IPR) are strengthened, then e may
decrease (while d can increase so that d + (n 1) e stays constant).
With only one type of technology, we cannot distinguish among innovation, develop-
ment, di¤usion, and learning by doing. Thus, several interpretations of Ri are consistent
3
The additive form of the noise is standard in the literature on di¤erential games (Başar and Olsder,
1999: Section 6.4; Dockner et al., 2000, Section 8.2), particularly when using quadratic functions (Engw-
erda, 2005: Section 9.1). The additive form of qG G and the assumption that qG 2 (0; 1) are particularly
natural in pollution settings (Dutta and Radner, 2004 and 2009).

5
with the model. For example, Ri may measure i’s abatement technology, i.e., the amount
by which i can at no cost reduce (or clean) its potential emissions. If energy production,
measured by yi , is generally polluting, the actual emission level of country i is given by:

gi = yi Ri . (2.3)

Alternatively, Ri may measure the capacity of country i’s renewable energy sources (or
"windmills"). If the windmills can generate Ri units of energy, and the alternative is to
use polluting fossil fuels, the total amount of energy produced is given by yi = gi + Ri )
(2.3).
Relying on (2.3), rather than focusing on technologies that reduce the emission content
of each produced unit (e.g., gi = yi =Ri ), simpli…es the analysis tremendously. An equally
helpful assumption is to let the investment cost be given by k (ri ). The additive form of
(2.3) has thus also been adopted elsewhere in the literature.4
Let the bene…t of consuming (and producing) energy be given by the increasing and
concave function B (yi ). If C (G) is an increasing convex function representing each coun-
try’s cost of the public bad, i’s utility in a period is:

ui = B (yi ) C (G) k (ri ) : (2.4)

With this, we are assuming a linearly separable utility function. It is important to recog-
nize the limitations of this, and to think about the consequences of relaxing this assump-
tion. Note, however, then when we abstract from uncertainty, maximizing ui is similar to
maximizing any monotonicly increasing function of ui . The results may thus be the same
if the real objective function is a Cobb-Douglas function:
e (yi )
B
;
e (G) e
C k (ri )
since, if we take the natural log of this equation, we can a linearly separable function
similar to (2.4).
Although we for some results can keep the functional forms in (2.4) general, we will
start by considering a linear investment cost, k (ri ) = Kri . In reality, the cost of investing
4
See, for example, Bos et al. (2014), analyzing how investment-contests can be designed to implement
the …rst–best. If the model focused instead on technologies that reduced the emission content of each
produced unit (as in Dutta and Radner, 2004), the analysis below would be much harder.

6
in technology can be a convex or a concave function (if there are increasing returns to
scale). Imposing linearity is thus a strong assumption, but it permits a tractable model
despite the n + 1 stocks.
In addition, we will be able to derive additional results by occassionally considering
the quadratic functional forms:

b c
B (yi ) = (y yi )2 and C (G) = G2 : (Q)
2 2
Here, parameters b > 0 and c > 0 measure the importance of energy and climate change.
The bliss point y represents the ideal energy level if there were no concern for pollution: no
country would produce more than y due to the implicit costs of generating, transporting,
and consuming energy. We can easily allow for heterogeneous bliss points y i ’s (see the
proofs).
Remarks on : The stochastic shock has a minor role in the model and most of the
results hold without it (i.e., if = 0). But the future marginal cost of emissions is in
reality uncertain, and this can be captured by . In fact, the model would be identical
b qG G + P gi and the uncertainty were in the
if the level of greenhouse gases were G i
b
associated cost-function, a¤ecting C but not G:

ui = B (yi ) b+
C G Kri ; where = qG + :

b and
Most results would continue to hold even if the e¤ects of G were not additive.5
Note that, although is i.i.d. across periods, it has a long-lasting impact through its
e¤ect on G.
Alternative interpretations: Instead of interpreting yi as "energy," we could substitute
(2.3) in B (:) and let B (gi + Ri ) measure i’s direct bene…t of adding to the public bad
(e.g., due to saved abatement costs). Better technology is then a perfect substitute for
producing the public bad. The public bad does not, of course, have to be greenhouse gases.
Moreover, by de…ning a public good as G and contributions as gi , i’s marginal cost of
providing the public good is B 0 (Ri ( gi )), increasing in i’s contribution but decreasing
in i’s technology. Hence, the model …ts many situations (with private provision of public
5
The exceptions are the results below that rely on quadratic utility functions.

7
goods or bads) beyond climate change.
To allow for alternative applications and to simplify the analysis, note that we have not
required gi or ri to be positive. Even for climate policies, a negative gi may be technically
feasible if carbon capture is an option, although this may be costly, as captured by a large
B 0 (:) for such a small gi . Similarly, ri may be negative if the abatement technology and
the associated infrastructure can be put to other uses.

2.2. Timing

The timing of dynamic games must be speci…ed. Here, it is natural to assume that the
investment stages and the pollution stages alternate over time. Somewhat arbitrarily, we
de…ne "a period" to be such that the countries …rst (simultaneously) invest in technology,
after which they (simultaneously) decide how much to pollute. In between, is realized.
Information is symmetric at all stages.
It is important for tractability that investments and emissions are decided on at dif-
ferent points in time.6 On the other hand, it is not important that the uncertainty be
realized between the investment stage and the emission stage, rather than vice versa. We
do not require the ri;t ’s or the gi;t ’s to be strictly positive.
Country i’s objective is to maximize the present-discounted value of its future utilities,
X
1
t
Ui = ui; ;
=t

where is the common discount factor and Ui is i’s continuation value as measured at
the start of period t. As mentioned, subscripts denoting period t are often skipped.

2.3. Equilibrium Concepts

There is typically a large number of subgame perfect equilibria in dynamic games, as in

repeated games. To be able to make more precise predictions, re…nements are necessary.
In dynamic and stochastic games, it is common to focus on Markov perfect equilibria
6
This assumption can be endogenized. Suppose the countries can invest at any time they want but
2 (0; 1) measures time required for the technology to mature or be built. Then, in equilibrium all
countries will invest at time t 2 (t 1; t).

8
 Figure 2.1: The investment and emission stages alternate over time.

(MPEs) in which strategies are conditioned only on the payo¤-relevant stocks (G and
R fR1 ; :::; Rn g). Note that this restriction would not make much sense in a repeated
game, since there there is no state or payo¤ relevant stock, so the MPE would simply be
the Nash equilibrium of the stage game.
There are several general reasons for selecting the Markov perfect equilibria. First,
Markov perfect strategies are simple, since they do not depend on the history in arbitrary
ways (Maskin and Tirole, 2001), which simpli…es the analysis as well. Second, experi-
mental evidence suggests that players tend toward Markov perfect strategies in complex
environments (Vespa, 2012; Battaglini et al., 2013). Third, focusing on the MPEs is
quite standard when studying games with stocks. Using an MPE in this note clari…es my
contribution to the literature.
In addition, Markov perfection is particularly attractive in the present model. In
contrast to much of the literature, there is a unique symmetric MPE in the present
game. This sharpens the predictions and makes institutional comparisons possible. The
restriction to symmetric MPEs means that if every country faces identical continuation
values at the investment stage, then we select the equilibrium where they invest the same
amount.7
This equilibrium coincides with the unique symmetric subgame perfect equilibrium if
time were …nite but approached in…nity.8 This is particularly important in the present
7
Since the investment cost is linear, there are asymmetric MPEs in which the countries invest di¤erent
amounts. In fact, if parameter b, c or k varied across countries, the MPE would have to be asymmetric
(they are thus analyzed in Harstad, 2012). In the present model, however, the asymmetric equilibria
would cease to exist if there were a slightly convex cost function for the investments ri;t . An additional
justi…cation is provided by Maskin and Tirole (1988:556), who also emphasize the symmetric MPE in
their model: "Our emphasis on symmetric equilibrium is meant to underscore the idea that the …rms are
inherently identical, so that, placed in the same circumstances, they should behave the same way".
8
This fact can easily be seen by the recursive nature of the proofs (see Harstad, 2012, for an explicit
proof). Fudenberg and Tirole (1991:533) suggest that "one might require in…nite-horizon MPE to be

9
context, since the equilibrium is then robust to the introduction of real-world aspects
that would make the e¤ective time horizon …nite. For example, since fossil fuel is an
exhaustible resource, the emission game may indeed have a …nite time horizon in the
real world. Similarly, politicians’ term-limits or short time horizon may force them to
view time as expiring. Finally, since the unique MPE makes it impossible to enforce
agreements by using trigger strategies, it becomes meaningful to focus instead on settings
where countries can negotiate and contract on emission levels–at least for the near future.
This note does not attempt to explain how countries can commit, but domestic rati-
…cation is seldom meaningless. In the United States, for example, the Supremacy Clause
(Article VI, paragraph 2 of the US Constitution) states that "all Treaties made... shall
be the supreme Law of the Land..." Thus, US states are bound to uphold a signed treaty,
even in the presence of con‡icting state laws. However, since nations’ability to commit
may in general be imperfect, I analyze alternative scenarios where countries cannot com-
mit at all (Section 3.3), where they can sign complete contracts (Section 3.2), or where
they contract on some but not all issues of interest (Sections 4-6). The last scenario turns
out to be most interesting analytically. This is also the scenario best describing current
climate negotiations. Note that I do not allow countries to commit to rules for how they
should negotiate in the future.9
At the negotiation stage, we will assume the bargaining outcome is e¢ cient and sym-
metric if it should happen that the game and the payo¤s are symmetric. This condition
is satis…ed whether we rely on (i) the Nash Bargaining Solution, with or without side
transfers, (ii) the Shapley value, or instead (iii) noncooperative bargaining in which one
country is randomly selected to make a take-it-or-leave-it o¤er specifying quotas and side
payments.
limits of …nite-horizon MPE."
9
For example, a commitment to uniform quotas could raise e¢ ciency for short-term agreements, and
would implement the …rst best if e = 0 and the uniform quota was determined by a majority requirement
short of unanimity.

10
3. General Analysis

This section solves the game above, gradually increasing the possibilities for negotiating
and contracting. Several contracting environments are analyzed in the following subsec-
tions:
Contract Complete None Incomplete + reneg.
One-period 3 5
1 2
Multiperiod 4 6
Naturally, di¤erent contracting environments generate di¤erent outcomes. Neverthe-
less, all the contracting environments share a number of attractive features. Rather than
repeating these features for each case, let us start by emphasizing them (although they
will be proven only later).

Theorem. For each contracting environment it is true that:

P
(1) The individual continuation value U is a function of only G and R N Ri; .
(2) Continuation values are linear in the stocks and satisfy:
@U (G ; R ) K
= qG (1 qR ) ;
@G n
@U (G ; R ) K
= qR :
@R n
(3) There is a unique symmetric MPE.
(4) This MPE coincides with the unique symmetric SPE if time is …nite, t 2 1; ::; t ,
and t < t 1.

Part (1) is easily illustrated. In principle, the continuation value Vi is a function of

the n + 1 stocks G and R [R1 ; R2 ; :::; Rn ]. However, note that choosing gi is equivalent
to choosing yi , once the Ri s are sunk. Thanks to the assumption that technology and
emissions are perfect substitutes, by combining (2.1) and (2.3), we get:
X
G = qG G + yi R.
N

This way, the Ri s are eliminated from the model: they are payo¤-irrelevant as long as R
is given, and i’s Markov perfect strategy for yi is therefore not conditioned on them.10
10
This follows from the de…nition by Maskin and Tirole (2001: 202), where Markov strategies are
measurable with respect to the coarsest partition of histories consistent with rationality.

11
Hence, a country’s continuation value Vi is a function of G and R , and not of Ri; Rj; ;
and we can write that value as V (G ; R ), without the subscript i.
The remaining parts of the theorem are best understood after the analysis of each
scenario and by following the proofs in the appendix. For example, in every scenario, the
value of another unit of technology is that each of the n countries can invest qR =n units
less in the following period: this explains the level @U=@R and why this is constant across
scenarios. The fact that U is linear in all stocks hinges on the linear investment costs,
revealing the power of this assumption. The linearity of V makes the model tractable and
permits only one equilibrium. To simplify notation, equations often rely on the conven-
tional de…nitions UG @U=@G , UR @U=@R , B 0 dB (yi ) =dyi , C 0 dC (G) =dG,
etc.

3.1. The …rst-best outcome

If investments as well as emissions were contractible, the countries would agree to the
…rst-best outcome. This follows from the observation (made in the previous subsection)
that the bargaining game is symmetric, even if the Ri s or the y i s di¤er. The outcome is
thus e¢ cient and coincides with the case in which a benevolent planner makes all decisions
in order to maximize the sum of utilities.

For future reference, we start by describing the …rst-best outcome, i.e., the outcome if
every gi;t and ri;t were contractible or chosen by a planner maximizing the sum of utilities
P
N Ui;t .

(1) The …rst-best consumption yi = gi + Ri is independent of Ri , given R:

B 0 (gi (R) + Ri ) n (C 0 UG ) = 0; (3.1)

(2) The …rst-best investment ensures that Ri satis…es:

B 0 (gi (R ) + Ri ) = (1 qR ) K: (3.2)

(3) The …rst-best pollution level G satis…es:

C 0 (G ) = (1 qG ) (1 qR ) K=n: (3.3)

12
 Part (1) simply states that, at the emission stage, the (private) marginal bene…t of
consumption should equal the (social) cost of pollution. Intuitively, this requires identical
consumption levels, yi = yj 8i; j 2 N 2 , regardless of the di¤erences in technologies.
Technological di¤erences do imply, however, that the ex post …rst-best emission level
should be smaller for i if Ri is large:

gi (R) yi Ri :

Part (2) states that the …rst-best investment level equalizes the marginal bene…t to the
marginal cost, recognizing that more investments today reduce the need to invest in the
next period. By substituting (3.2) in (3.1), we …nd the …rst-best pollution level, stated in
part (3). Interestingly, note that both yi and G are independent of the past stocks: G
P
is pinned down by (3.3) and, together with G , this pins down N gi . Since yi is pinned
down by (3.2), Ri = yi gi follows as a residual, and so does ri = Ri qR Ri; .
Expectations are w.r.t. the unknown realization of . Combined with (2.1), (3.3) pins
P
down the i gi s. Given the gi s, (3.2) determines the …rst-best Ri s which, with (2.2),
determine the …rst-best ri s. Throughout the analysis, we assume gi 0 and ri 0 never
bind.11
Note that, under (Q), the equations above can give us explicit solutions:

(i) The …rst-best emission levels are functions of the technology stocks, R fR1 ; :::; Rn g:
cn (ny + qG G + R) + qG (1 qR ) K
gi (R) = y i Ri . (3.4)
b + cn2
(ii) The symmetric …rst-best investments are:
qR qG 1 qG 1
ri = y R + G (1 qR ) + K:
n n cn2 b
(iii) Combined, the …rst-best pollution stock is:
X (1 qG ) (1 qR ) b
G = gi (R ) + qG G = K+ : (3.5)
i2N
cn b + cn2

The dynamics are noteworthy. If the random pollution shock t happens to be large,
the emissions are reduced somewhat but the total stock, Gt , is nevertheless increasing in
11
This is satis…ed if gi < 0 and ri < 0 are allowed, or if qG and qR are su¢ ciently small.

13
 t. In the following period, the countries’investments are so much larger, and the optimal
emission levels are thus so much smaller, that the stock Gt+1 returns to the original level
(independent of t ). Since the large technology stock survives to period t + 2, investments
are then reduced. The steady state is reached after two periods–thanks to the linear
investment cost:12

Corollary 1. In the …rst-best outcome, a large t reduces gt and gt+1 but raises rt+1 :

@gt @gt+1 @rt+1 1 @rt+2

= and = qG (1 )= = , where
@ t @ t @ t qR @ t
cn
2 (0; 1) :
b + cn2

3.2. Business as Usual

In the other extreme scenario, neither emissions nor investments are negotiated. This may
be reasonable if the countries cannot commit to future policies because e¤ective sanctions
are lacking. This noncooperative situation is referred to as "business as usual."
Note that choosing gi is equivalent to choosing yi , once the Ri s are sunk. Substituting
(2.3) into (2.1), we get:
X
G = qG G + + yi R, (3.6)
i
X X
R Ri = qR R + ri D. (3.7)
i i

This way, the Ri s are eliminated from the model: they are payo¤-irrelevant as long as
R is given, and i’s Markov Perfect strategy for yi is thus not conditioned on them.13 A
country’s continuation value Ui is thus a function of G and R , not Ri; Rj; , and we
can therefore write it as U (G ; R ), without the subscript i.
12
With linear contribution costs, it is a typical result that an MPE can reach the steady state in a
single step; see Dockner et al. (1996). Also my model would feature a one-step transition with shocks
in the technology stock. However, as described by my corollaries, with a pollution stock (and nonlinear
emission bene…ts) as well as technology stocks, the transition lasts two periods when the shock regards
the pollution. If investment costs were convex, the transition would be more gradual (and reasonable).
13
This follows from the de…nition by Maskin and Tirole (2001, p. 202), where Markov strategies are
measurable with respect to the coarsest partition of histories consistent with rationality.

14
 At the emission stage, when the technologies are sunk, i solves

max B (yi ) C (G) + U (G; R) s.t. (3.6) )

yi

B0 C 0 + UG = 0: (3.8)

First, note that, since (3.8) decreases in gi , each country pollutes too much compared
to the …rst best (3.4) when UG < 0: a country is not internalizing the cost for the others.
Second, (3.8) con…rms that each i chooses the same yi , no matter the Ri s. While
perhaps surprising at …rst, the intuition is straightforward. Every country has the same
preference (and marginal utility) w.r.t. yi , and the marginal impact on G is also the same
for every country: one more energy unit generates one unit of emissions.14
Substituting (3.6) in (3.8) implies that a larger R must increase every yi . This implies
that if Ri increases but Rj , j 6= i, is constant, then gj = yj Rj must increase. Further-
more, substituting (2.3) in (3.8) implies that if Ri increases, gi must decrease. In sum: if
country i has better technology, i pollutes less but (because of this) other countries pollute
more. Clearly, this e¤ect discourages countries from investing. In addition, the Appendix
shows that, in equilibrium, ri increases in G but decreases in R . Thus, a country may
want to pollute a lot and invest little today in order to induce the other countries to invest
more tomorrow. With these dynamic e¤ects, this common pool problem is more severe
than its static counterpart (or the open-loop equilibrium).15

Proposition 1: There is a unique symmetric MPE. Countries pollute too much and
invest too little. Furthermore:

yino = yjno 8i; j 2 f1; :::ng8Ri ; Rj ;

@gino =@Ri < 0 < @gino =@Rj 8j 6= i;

@rino =@G = qG =Dn, @rino =@R = qR =Dn,

URno = qR K=Dn, UGno = qG (1 qR ) K=Dn: (3.9)

14
Obviously, this would not necessarily be true if I instead had focused on technologies that reduce the
emission factor of each produced unit (as do Dutta and Radner, 2004). The additive form (2.3) is chosen
not only because it simpli…es the analysis tremendously, but also because the resulting crowding-out
e¤ects might be reasonable in reality.
15
This is also the case in Ploeg and de Zeeuw (1991), for example.

15
 Conveniently, the continuation value U is linear in G and R : Thus, the n + 1
stocks collapse to one, making the analysis tractable. This is thanks to (2.3) and the
linear investment cost, which also ensures that the equilibrium is unique.16 Note that the
equilibrium is also stationary.
The dynamic paths are simple. Following a large , every country pollutes less and,
in the next period, investments increase such that G+ returns to the original level. The
steady state is thus reached in one period.
The size of the externality e has no e¤ect, given D. Since only R matters for utilities
and strategies, Ri becomes a pure public good.17
For a given R, countries pollute more if qR is large and K=D small. Intuitively, if the
technology is e¢ cient, cheap, and long-lasting, one can pollute more today and let this
motivate investments in technology tomorrow. This, and other comparative statics can
be observed in the Appendix, which derives explicit formulae for the case where B (:) and
C (:) are quadratic functions.
We can get more speci…c solutions. By di¤erentiating (3.8) and abstracting from
uncertainty:

(1) The equilibrium consumption yino is independent of Ri and suboptimally large,

given R, :
B 0 (yino ) = C 0 UG : (3.10)

(2) Country i pollutes less but j 6= i pollutes more if Ri is larger, …xing Rj 8j 6= i:

C 00 (n 1) B 00
@gino =@Ri = < 0;
nC 00 B 00
C 00
@gjno =@Ri = > 0 8j 6= i;
nC 00 B 00
16
As the proposition states, this is the unique symmetric MPE. Since the investment cost is linear,
there also exist asymmetric MPEs in which the countries invest di¤erent amounts. Asymmetric equilibria
may not be reasonable when countries are homogeneous, and they would cease to exist if the investment
cost were convex. Multiple equilibria never arise under long-term agreements.
17
Thus, adding to the public good G (by reducing gi ) or to R (by increasing ri ) has somewhat
similar e¤ects. However, they are not equivalent since a larger ri reduces emissions in every future
period. Increasing ri thus has longer-lasting impact than reducing gi , which is why ri is referred to as an
investment. Moreover, the remainder of this section lets gi be contractible while ri is not.

16
 (3) Equilibrium investments are suboptimally low and ensure that:

0 nC 00 B 00
B (Rino + gino ) = (1 qR =n) K; (3.11)
C 00 B 00

assuming that the following second-order condition holds:

" #
000 00 2 00 000 00
B (C ) =B C B
C 00 (n 1) B0: (3.12)
(C 00 B 00 ) (nC 00 B 00 )

(4) Every country invests more if R is small and G large:

@rino =@R = qR =n;

@rino =@G = qG =n:

(5) The resulting pollution level is suboptimally large and given by:

(n qR ) C 00
C 0 (Gno ) = (1 qG ) (1 qR ) K=n + 1 + (1 1=n) K (3.13)
C 00 B 00

Compared to the …rst-best (3.2), the right-hand side of (7.9) is larger. This implies that
country i’s consumption, Rino + gi , is smaller than at the …rst-best. Since Proposition 2
states that consumption levels are suboptimally large, given R, it follows that investments
must be suboptimally small: gino (Rno ) + Rino < gi (R ) + Ri and gi (Rno ) > gi (Rno )
implies Rno < R . Furthermore, the e¤ect of suboptimally small investments dominate
the suboptimally large emissions, leading to lower equilibrium consumption than at the
…rst-best.18
Part (2) …rst states that, if the technology stock is small today, every country invests
more in the next period. Naturally, this contributes to explaining the underinvestment
result. Second, if the pollution level is large, all countries invest more. This contributes
to explaining the suboptimally large emissions, as described by Proposition 2. Part (3)
states that, when all these e¤ects are taken into account, the resulting pollution level is
larger than the …rst-best level.
18
Note the qualifying second-order condition, which is violated if C 000 is positive and large and/or if
000
B is very negative. In both situations, an increase in R raises yj at a decreasing rate. Thus, the
discouragement e¤ect when i invests (i.e., the fact that other countries pollute more) is reduced when i
invests more, and this implies that an interior solution for ri may not exist. However, the second-order
condition holds if C 000 and B 000 are small relative to C 00 and B 00 ; and it always holds for quadratic
functions (implying C 000 = B 000 = 0).

17
 Figure 3.1: The timing for short-term agreements.

With (Q), we get explicit solutions:

With (Q) under business as usual, countries pollute too much and invest too little:

qR qG
ribau = y R + G
" n n #
2
(b + cn) qR qG
e (n 1) + 1 K (1 qR ) 2 K
cb (b + c) n n cn
< ri ;
c (ny + qG G + R) + qG (1 qR ) K=n
gibau Rbau = yi Ri
b + cn
> gi Rbau > gi (R ) : (3.14)

Corollary 2. With business as usual, Corollary 1 continues to hold if is replaced by

bau
= c= (b + cn) < .

3.3. Incomplete Contracts and Hold-Up Problems

If countries can commit to the immediate but not the distant future, they may negotiate
a "short-term agreement." If the agreement is truly short-term, it is di¢ cult for the coun-
tries to develop new technology during the time-span of the agreement and the relevant
technology is given by historic investments. This interpretation of short-term agreements
can be captured by the timing of Figure 2.
Negotiating the gi s is equivalent to negotiating the yi s as long as the Ri s are sunk and
observable (even if they are not veri…able). Just as in the previous section, (3.6)-(3.7)
imply that the Ri s are payo¤-irrelevant, given R. Even if countries have di¤erent Ri s,

18
they face the same marginal bene…ts and costs of yi whether negotiations succeed or not.
Symmetry thus implies that yi is the same for every country in the bargaining outcome.
E¢ ciency implies that the yi s are optimal (all countries agree on this):

B0 nC 0 + n UG = 0 ) (3.15)

gist = gi ;

where both gi and gist are functions of existing technology and pollution levels.
Substituting (3.6) in (3.15) and (2.3) in (3.15) implies that if Ri increases, gi must
decrease - but gj increases, 8j 6= i. Intuitively, if i has better technology, i’s marginal
bene…t from polluting is less (and i is also polluting less in equilibrium). This gives i a
poor bargaining position, and the other countries can o¤er i a smaller emission quota.
At the same time, the other countries negotiate larger quotas for themselves, since the
smaller gi (and the smaller G) reduce the marginal cost of polluting. Anticipating this
hold-up problem, every country is discouraged from investing. The Appendix shows i
invests until
K (n qR )
EB 0 gi + Rist = ; (3.16)
D
so Rist is smaller than the optimal one, given by (3.2). The equilibrium pollution level is

(1 qG ) (1 qR ) K K (1 1=n)
EC 0 (G) = + :
Dn D

Thus, although emission levels are ex post optimal (3.15), once the investments are sunk,
G is larger compared to its …rst-best level (3.3) since the hold-up problem discourages
investments and makes it ex post optimal to pollute more.

Proposition 2: Proposition 1 continues to hold: There is a unique symmetric MPE.

Countries pollute too much and invest too little. Furthermore:

yist = yjst 8i; j 2 f1; :::ng8Ri ; Rj ;

@gist =@Ri < 0 < @gist =@Rj 8j 6= i;

@rist =@G = qG =Dn, @rist =@R = qR =Dn,

URst = qR K=Dn, UGst = qG (1 qR ) K=Dn:

19
 While its intuition is quite di¤erent, Proposition 2 is identical to Proposition 1. In
particular, UG and UR are exactly the same as in the noncooperative case. This does not
imply that U itself is identical in the two cases: its level can be di¤erent. But this does
imply that in deriving actions and utilities for one period, it is irrelevant whether there
will also be a short-term agreement in the next (or any future) period. This makes it
convenient to compare short-term agreements to no agreement.

3.3.1. Are Short-Term Agreements Good?

Pollution is less under short-term agreements compared to no agreement. That may not
be surprising, since the very motivation for negotiating is to reduce pollution. But what
about equilibrium investments and utilities?
Unfortunately, a general comparison is not feasible. But some insight can be generated
by assuming B 00 (:) and C 00 (:) are constants as in Q.

Proposition 3: Under (Q), short-term agreements reduce (i) pollution, (ii) investments,
and (iii) utilities if n is large and each period short (i.e., if (3.17) holds):

K n 1 qR
EGst = EGno 1 ;
D b+c n
K (n 1)2 qR
rist = rino 1 ;
nD2 (b + c) n
" #
(b + c) (bcnD=K)2
(n 1)2 (1 q R )2 > 2
: (3.17)
(b + cn2 ) (b + cn)2

Rather than being encouraging, short-term agreements impair the motivation to invest.
The reason is the following. Anticipating negotiations, the hold-up problem is exactly as
strong as the crowding-out problem in the noncooperative equilibrium: in either case, each
country only enjoys 1=n of the total bene…ts generated by its investments (no matter e).
In addition, when an agreement is expected, i understands that pollution will be reduced.
A further decline in emissions, made possible by new technology, is then less valuable.
Hence, each country invests less.19
19
A counter-argument is that, if an agreement is expected, it becomes more important to invest to
ensure a decent energy consumption level. While this force is smaller under (Q), it could dominate for
other functional forms.

20
 Since investments decrease under short-term agreements, utilities can decrease as well.
This is the case, in particular, if the period for which the agreement lasts is truly short.
If so, and qR are large, while there is not much uncertainty from one period to the
next. All changes make (3.17) reasonable, and it always holds when the agreement is very
short ( ! 0). Moreover, (3.17) is more likely to hold if n is large (it always holds if
n ! 1): the under-investment problem is then large, it becomes important to increase
investments, and this is achieved by having no agreement.
At the emission stage, however, once the investments are sunk, all countries bene…t
from negotiating an agreement. It is the anticipation of negotiations which reduces in-
vestments and perhaps utility. Thus, if (3.17) holds, the countries would have been better
o¤ if they could commit to not negotiate short-term agreements. In particular, it may be
better to commit to emission levels before the investments occur.

3.4. Optimal Contracts

The model can (and will) be used to analyze agreements of any length. If the countries can
negotiate and commit to future emission levels, it will be possible to develop technologies
within the time-frame of an agreement. The other countries are then unable to hold up
the investing country, since the quotas have already been negotiated, at least for the near
future.

3.4.1. One-period Agreements

This interpretation of "long-term agreements" can be captured simply by letting the

countries negotiate the gi s in the beginning of each period, before the investments are
made. While these agreements last only one period, they are indeed "longer" than the
short-term agreements studied in Section 4.2. Moreover, each period can be arbitrarily
long in the model, since I have not speci…ed whether the discount factor, for example, is
large or small.
For each period, the timing is now reversed. When investing, a country prefers a larger
stock of technology if its quota, gilt , is small, since otherwise it is going to be very costly

21
 Figure 3.2: The timing for long-term agreements.

to comply. Consequently, ri decreases in gilt . The Appendix shows that ri increases until
K (1 qR =n)
B 0 gilt + Rilt = : (3.18)
D e (n 1)
Compared to (3.16), (3.18) suggests that countries invest more under long-term than
under short-term agreements (at least for the same gi ). But compared to the …rst best
(3.2), countries still under-invest if e > 0 or qR > 0. First, a country does not internalize
the spillover e on the other countries. Second, if the agreement does not last forever
( > 0), a country anticipates that good technology worsen its bargaining position in the
future, once a new agreement is to be negotiated. At that stage, good technology leads
lt
to a lower gi;+ since the other countries can hold up i when it is cheap for i to reduce its
emissions.20 This discourages i from investing now, particularly if the current agreement
is relatively short ( large) and the technology likely to survive (qR large). In sum, if e,
, and qR are large, it is important to encourage more investments. This can be achieved
by a small gilt .
The Appendix shows that the equilibrium and optimal gilt s must satisfy (3.3): the
equilibrium pollution level is similar to the …rst best! But since (3.18) implies that the
equilibrium Rilt s are less than optimal, the gilt s are suboptimally low ex post. Combining
(3.3) and (3.18):
K e(1 qR )(n 1) + qR (1 1=n)
B0 EnC 0 n UG = ) (3.19)
D D e (n 1)
K e(1 qR )(n 1) + qR (1 1=n)
gilt = Egi if (Q).
D (b + cn2 ) D e (n 1)
20
Or, if no agreement is expected in the future, a large Ri;+ reduces gi;+ and increases gj;+ , as proven
in Section 4.1.

22
Taking the investments as given, optimally the gilt should have satis…ed B 0 EnC 0
n UG = 0 rather than (3.19). Only that would equalize marginal costs and bene…ts
of abatement. Relative to this ex post optimal level, the gilt satisfying (3.19) must be
lower. If e and are large, the right-hand side of (3.19) is large, and gi must decline.
This makes the long-term agreement more demanding or tougher to satisfy at the emis-
sion stage. The purpose of such an overambitious agreement is to encourage investments,
since these are suboptimally low when e and are large.

Proposition 4: (i) There is a unique MPE. (ii) Each country invests more if the agree-
ment is tough (3.18). Therefore, (iii) the optimal agreement (3.19) is tougher if the
externality e is large and the time horizon short ( large).

On the other hand, if e = qR = 0, the right-hand side of (3.19) is zero, meaning that
the commitments under the best long-term agreement also maximize the expected utility
ex post. In this case, there are no externalities, and the countries are not concerned with
how current technologies a¤ect future bargaining power. Thus, investments are …rst best
and there is no need to distort the gilt s downwards.
The continuation value U is linear in the stocks, making the analysis tractable. More-
over, URlt = qR K=Dn and UGlt = qG (1 qR ) K=Dn, just as in the two previous subsec-
tions. The predicted contract and investments are therefore robust to whether there is a
long-term agreement, a short-term agreement, or no agreement in the subsequent period.

3.4.2. Multiperiod Agreements

Assume now that at the beginning of period 1, countries negotiate the gi;t s for every
period t 2 f1; 2; :::; T g. When investing in period t 2 f1; 2; :::; T g, countries take the gi;t s
as given, and the continuation value in period T + 1 is U (GT ; RT ). At the last investment
stage, i’s problem is the same as before and i invests until (3.18) holds. Anticipating this,
i can invest less in period T by investing more in period T 1. The net investment cost
is thus K (1 qR ). The same logic applies to every previous period and, in equilibrium,

K (1 qR ) K (1 qR )
EB 0 (gi;t + Ri;t ) = = for t < T: (3.20)
d D e (n 1)

23
 Thus, the incentives to invest are larger earlier than in the last period (3.18). In fact,
if e = 0, investments are …rst best for every t < T . In the last period, however, countries
invest less, anticipating the hold-up problem in period T + 1.21
All this is anticipated when the countries negotiate the gi;t s. As shown in the Appendix,
the optimal gi;t s must satisfy (3.3) for every t T : the pollution level is similar to the
…rst best! The gi;t s are thus lower than what is optimal ex post when e > 0 and countries
under-invest. Combining (3.3) and (3.20) for t < T ,

K e (n 1) (1 qR )
B0 EnC 0 n UG = ) (3.21)
D D e (n 1)
K e (n 1) (1 qR )
gi;t = Egi if (Q).
D (b + cn2 ) D e (n 1)

Ex post, B 0 nC 0 n UG = 0 is optimal. Compared to this, gi;t satisfying (3.21) should

be smaller if e is positive and large. For t = T , however, (3.19) continues to hold and
since its right-hand side is less than that of (3.21), yi;T < yi;t for t < T . In words: in order
to encourage investments, the agreement should be tougher to satisfy toward the end.

Proposition 5: Suppose countries negotiate emission levels for T periods. (i) There is
a unique MPE. (ii) Investments decrease toward the end and, to encourage more invest-
ments, (iii) the equilibrium agreement becomes tougher over time compared to the ex post
optimum (3.19)-(3.21).

3.4.3. The Optimal Length of an Agreement

The optimal T trades o¤ two concerns. On the one hand, investments are particularly
low just before a new agreement is to be negotiated. This hold-up problem arises less
frequently if T is large. On the other hand, the stochastic makes it hard to predict the
optimal gi;t s for the future, particularly when T is large. If were known or contractible,
the agreement should last forever. Otherwise, one can show that the optimal T declines
as e # 0. The Appendix derives a large number of comparative statics for the case where
B and C are quadratic (Q):
21
Or, if no agreement is expected in period T + 1, i anticipates @gj;T +1 =@Ri > 0, j 6= i.

24
Proposition 6: (i) Under (Q), the optimal length is …nite, T < 1, if and only if:
2 2
K qR K bcqG
e+ e [2 qR ] + < :
n n qR (1 2
qG ) (1 qG2
) (n 1)2

(ii) Under this condition, T increases in e, n, qR , and K, but decreases in b, c, and .

Intuitively, the under-investment problem is particularly severe if e and n are large.

Reinforcing this problem by a small T is then especially harmful, and the optimal T is
larger. Naturally, T should be smaller if future optimal emissions are uncertain ( large)
and important (c large).22

3.5. Renegotiation and Renegotiation Design

The long-term agreements above are not renegotiation-proof. Not only are the commit-
ments made before the severity of the problem (determined by ) is known, but they also
specify emission levels that are less than what is expected to be optimal ex post. The
countries may thus be tempted to renegotiate the treaty, after and the investments are
realized. This section derives equilibria when renegotiation is costless.

3.5.1. One-period Agreements and Renegotiation

The timing in each period is now the following. First, the countries negotiate the initial
commitments, the gide s, referred to as "the default." If these negotiations fail, it is natural
to assume that the threat point is no agreement.23 Thereafter, the countries invest and
is realized. Before carrying out their commitments, the countries get together and
renegotiate the gide s. Relative to the threat point gide , the bargaining surplus is split
equally.24
Renegotiation ensures that emission levels are ex post optimal, in contrast to the long-
term agreements in Section 4.3. When investing, a country anticipates that it will not, in
22
If b is large, consuming energy is much more important than the concern for future bargaining power,
the hold-up problem vanishes, and the optimal T is smaller.
23
If the threat point were short-term agreements, negotiated after the investment stage, the outcome
would be identical.
24
If instead the threat point at the renegotiation stage were that the countries would get upset and
revert to no cooperation, the renegotiation game would be identical to negotiations under short-term
agreements, and the incentives to invest would be as discussed in Section 4.2.

25
the end, have to comply with an overambitious long-term agreement. Will this jeopardize
the incentives to invest?

Proposition 7: (i) There is a unique MPE. (ii) The initial agreement satis…es (3.22):
the initial quota gide is thus smaller if the spillover e is large and the time horizon short
( large). (iii) All investments and emissions are …rst best.
K
B 0 gide + Ri = ) (3.22)
D en
K en
gide = Egi + qR under (Q).
bD D en
When investing, the countries do anticipate that, after renegotiation, emissions will
be ex post optimal, just as they were under a short-term agreement. But for the short-
term agreement, countries with the poorest technology got the better deal, since these
countries were quite satis…ed with the (noncooperative) default outcome in which they
could pollute more. This made the "technology-losers" reluctant to negotiate, giving them
a better bargaining position. However, things are quite di¤erent when renegotiating an
ambitious agreement. Then, the technology-losers are desperate to reach a new agreement
that would replace the initial commitments. Such countries now have a poor bargaining
position, and they are, in equilibrium, going to get quite a bad deal (where they must pay
or accept a small gi ). Fearing this, the countries are induced to invest more, particularly
if the default emission levels are small.
All this will be taken into account when negotiating the initial agreement, the gide s.
The more ambitious this agreement is, the more the countries invest. This is desirable if
the countries are otherwise tempted to under-invest. Thus, the agreement should be more
ambitious if e and qR are large. Formally, (3.22) implies that gide decreases in e and qR

26
since Ri is increasing in qR but independent of e. Intuitively, if the length of the agreement
is short, countries fear that more technology today will hurt their bargaining position in
the near future. They thus invest less than what is optimal, unless the agreement is more
ambitious.25
Compared to (3.19), the initial agreement should be tougher than the optimal long-
term agreement (gide < gilt ). Intuitively, the long-term agreement (without renegotiation)
balances the concern for investments (by reducing gilt ) and for ex post e¢ ciency (in which
gi should be larger). The latter concern is irrelevant when renegotiation ensures ex post
optimality, so the initial contract can be tougher - indeed so tough that investments are
…rst best.

3.5.2. Multiple Periods and Renegotiation

The …rst best is implemented by any long-term agreement lasting T 1 periods if rene-
de
gotiation is possible. Suppose an agreement speci…es gi;t , t 2 f1; :::; T g. Investments at
t < T are …rst best if:

K (1 qR ) K 1 qR
de
B 0 gi;t + Ri = de
) gi;t = Egi;t if (Q). (3.23)
D en bD D=en 1
de
Compared to (3.22), gi;t is larger when T > 1 than when T = 1 (Ri is independent
of T ). Thus, agreements lasting one period should be more ambitious than if T > 1,
con…rming the earlier …nding that an agreement should be more ambitious if its length is
short.

Proposition 8: Suppose countries negotiate emission levels for T > 1 periods and rene-
de
gotiation is possible. At t < T , all investment and emission levels are …rst best if gi;t is
de 0 26
given by (3.23). T and gi;t 0 , t > t, are irrelevant.

25
No contract can help if the externality dominates the direct e¤ect (e d), as …rst pointed out by
Che and Hausch (1999) and later generalized by Segal and Whinston (2002).
26 de 0
Since T and gi;t 0 , t > t, are irrelevant, the predictions are not sharp when renegotiation is feasible.
de
With a small …xed cost of negotiating each gi;t 0 , however, the unique optimal contract would be described

by Proposition 7.

27
3.5.3. Implementation

de
The optimal gi;t+1 s depend on t and they must be (re)negotiated after t is realized. But
de
instead of being negotiated at the start of period t + 1, the gi;t+1 s may equally well be
negotiated just before the emission stage in period t, since no information or individual
decisions are made in between. At this time, therefore, the countries may negotiate every
de de
gi;t+1 <Egi;t+1 while simultaneously renegotiating the gi;t s and replacing them by the
de
optimal gi;t+1 s, which are expected to be larger than the gi;t s negotiated in advance. This
might be observationally equivalent to a time-inconsistent policy where the countries make
ambitious plans for the future, while repeatedly backing down from promises made in the
past. But rather than verifying a time-inconsistency problem, this leads to the …rst best.

Corollary 3: In equilibrium, the countries repeatedly promise to pollute little in the

future but when the future arrives, they relax these promises. This procedure implements
the …rst best.

4. Trade, Tari¤s and Intellectual Property Rights

This section introduces a new externality, relates it to trade agreements and analyzes the
e¤ects of and for climate treaties.
Externalities: There are several ways in which spillovers could be formalized, but
many of them give similar results to those above.27 An alternative (or addition) to the
spillover above arises by assuming that j bene…ts directly by the externality xri when
i 6= j invests. If K continues to be the social net marginal investment cost, i’s private
investment cost is
k K + (n 1) x: (4.1)

The model is unchanged if we just write the utility as

X
ui = B (yi ) C (G) kri + xrj :
j6=i

27
For example, the spillover could be related to Ri rather than ri (as in Coe and Helpman, 1995). The
results would be similar, but i must then consider the impact of ri on Rj not only for the present, but
for all future periods.

28
 The externality x can be interpreted as a general technological spillover (a¤ecting ui
and not only i’s environmental technology) or as a spillover that reduces i’s cost of making
P
a particular investment ri (such that the cost is kri xrj ).28 The Appendix allows for
both e and x and …nds them to play similar roles.
Trade in technologies: The externality x may re‡ect international law. Suppose
that ri has the potential of reducing j’s cost (or increasing uj ) by xri units. Of this, j can
copy a fraction 2 [0; 1] for free. The remaining fraction, 1 , is available if j pays i
for transferring (or licensing) its technology. If i sets the price, i charges j’s willingness to
pay, (1 ) x, for each invested unit. Clearly, the net externality for j is x = x. Weaker
intellectual property rights mean larger , x and k.
Firms: Private …rms have been ignored so far in the analysis. But since …rms may
develop most of the technology in reality, it is comforting to note that the results would
not necessarily change if …rms were introduced. If the government can perfectly regulate
the …rms’investments in technologies by specifying conditional fees or grants, then …rms
are perfect agents for the government and it is su¢ cient to consider the government’s in-
centives. Even without regulation, if the governments are outsourcing the development of
technology and …rms compete by setting prices, anticipating the revenues (n 1) (1 )x
when licensing to foreigners, technological units are provided at cost-price k and …rms are
not a¤ecting the game.
Tari¤s and subsidies: If the foreign individuals or …rms paying for the externality
(1 ) x face an ad valorem tari¤ , they are willing to pay only (1 ) (1 ) x for each
imported unit. On the other hand, since trade is veri…able, one may consider encouraging
R&D by subsidizing trade in abatement technology.29 Let s represent this subsidy, paid
for by either the importing country or uniformly by the non-exporting countries. In either
case, the net externality for country j when country i decides to invest becomes

x= x (s ) (1 ) x:

The private cost of investing faced by a country (or government) is k given by (4.1), just
28
In fact, the cost would take exactly this form if countries simultaneously choose their targets for the
Ri s and let the expenditures (the ri s) follow residually from (2.2) rather than vice versa.
29
Stern (2007, p. 398) states "There are two types of policy response to spillovers... enforcement of
private property rights through patenting [and] government funding."

29
as before. Thus, k and x are both higher with tari¤s, since the importing country is then
capturing more of the surplus, but lower if importers subsidize technological trade. The
role of s would be identical if directed to investments rather than the associated trade.

Proposition 9: If the subsidy ( s) is low, tari¤ ( ) high and the intellectual property
rights weak ( large), (i) the agreement should be tougher and more long-lasting while (ii)
short-term agreements are likely to be worse than no agreement under (Q).

While the proof is in the Appendix, the intuition is straightforward. With tari¤s, small
subsidies and weak property right protection, …rms do not capture the bene…t experienced
by the foreigners. This forces the government to pay more to compensate the …rms when
investing, and they invest less. A further reduction in investment is then particularly bad,
making short-term agreements worse than the noncooperative equilibrium (under (Q)).
To encourage more investments, it is better to negotiate an agreement that is tougher and
more long-lasting.
If s, or can be speci…ed by international law, one may ask what their levels should
be. Does the optimal subsidy, tari¤ and intellectual property right protection depend on
the climate treaty?

Proposition 10: s should be larger while and smaller if the agreement is short-
lasting. The optimal s, and are given by (4.2) for short-term agreements, (4.3) for
long-term agreements (and the last period of multiperiod agreements), and by (4.4) for
multiperiod agreements (except for the last period).
K
sst st
1 st st
= > (4.2)
nx
K en
slt lt
1 lt lt
= qR + (1 qR ) > (4.3)
nx D
Ke
st t
1 t t
= : (4.4)
xD
It is more important to encourage investments by protecting intellectual property
rights, subsidizing technological trade and reducing tari¤s if the climate treaty is short-
lasting, since the hold-up problem is then larger. Such "trade agreements" are thus
strategic substitutes to climate treaties: Weakening cooperation on one area makes further
cooperation on the other more important.

30
 If the subsidy can be freely chosen, short-term agreements are …rst best: while (4.2)
induces optimal investments, countries are negotiating the ex post optimal emissions. The
emission levels under long-term agreements (without renegotiation) are never …rst best,
however, due to the stochastic .

Corollary 4: If s can be freely chosen, short-term agreements are …rst best while long-
term agreements (without renegotiation) are not.

If renegotiation is possible, the …rst best is feasible no matter s, and as long as the
initial agreement is more ambitious for large e and x (and thus large and but small
s). Under (Q), the gide s should satisfy30

K x=K + e=D
gide = Egi + qR : (4.5)
bD 1=n e=D

If the gide s are exogenously given and high (e.g. because a tough climate treaty would
be impossible to enforce), e¢ ciency and (4.5) are still ful…lled if just x is su¢ ciently
small (requiring a large s and small or ). This suggests that less ambitious climate
treaties should be accompanied by technological subsidies, low tari¤s and property right
protection, con…rming that the two types of agreements are strategic substitutes.

Corollary 5: With renegotiation, the …rst best is implemented even if gide increases, if
just s increases or or decreases.

5. Generalizations and Extensions

The analysis in these notes hinges on a number of strong assumptions. It is important to

think through these, and re‡ect on how the results may change if one instead make other
assumptions.
At the same time, a few assumptions as quite innocent, and can easily be relaxed. We
may consider two of these that are particularly relevant.
30
The general conditions are derived in the Appendix.

31
5.1. Tradable Permits

To simplify intuition and the reasoning, it has been assumed that quotas cannot be traded.
This is not a critical assumption, however.

Proposition 11: (i) All results survive with tradable permits, no matter whether side
payments are available. (ii) The equilibrium and optimal permit price is B 0 , thus increas-
ing in e and larger if T = 1 than if T > 1.

B 0 (gi + Ri ) is the value of being allowed to pollute one more unit, keeping G and R
constant. Proposition 11 follows by noting that, …rst, there is never any trade in permits
in equilibrium. Hence, if i invests as above, the marginal bene…t of more technology is
the same. Second, if i deviated by investing more (less), it’s marginal utility of a higher
technology decreases (increases) not only when permit-trade is prohibited, but also when
trade is allowed since more (less) technology decreases (increases) the demand for permits
and thus the equilibrium price. Hence, such a deviation is not attractive. When permits
are tradable, altering their allocation is a form of side transfer, making the feasibility of
explicit transfers irrelevant.31;32

5.2. Heterogeneity

So far, countries have been completely symmetric and there has been no heterogeneity. It
did turn out, however, that for a given R , di¤erences in Ri; (such as Ri; Rj; ) were
payo¤-irrelevant. It is therefore not necessary to assume that all countries start out with
the same technology.
Moreover, since the continuation values are linear in R, countries are risk-neutral in
that it would not matter if qR were random, as long as the expected depreciation rate is
1 qR . The realized depreciation can also be di¤erent for every country, as long as the
expected depreciation rate is 1 qR for everyone.
31
In a two-stage model, also Golombek and Hoel (2005) …nd that the permit price should be higher
than "the Pigouvian" level to induce R&D when there are spillovers.
32
An earlier version of this paper analyzed emission taxes and derived similar results: for example, the
…rst best is feasible with renegotiation if the initial tax is higher than what is expected to be optimal ex
post, particularly if the spillover is large and the agreement’s length relatively short.

32
 A strong assumption has been that all countries had identical preferences. With
quadratic utility functions, for example, it is reasonable to assume that countries have
di¤erent bliss points (y i ) for energy consumption. Generalizing the quadratic speci…cation,
we may write the bene…t function as B (yi y i ), where y i is a country-speci…c reference
point (not necessarily bliss). Recognizing the importance of such heterogeneity, all proofs
allow the reference point y i to vary. While a large y i increases the equilibrium gi , the
comparative statics are unchanged.33

Proposition 12: All results continue to hold with heterogenous bliss or reference points
in B (yi y i ).

6. Further Readings

These lectures notes mostly draw on Harstad (2012 and 2015). But by analyzing environ-
mental agreements as incomplete contracts in a dynamic game, we relate to three strands
of literature.
The literature on climate policy and environmental agreements is growing.34 While
Nordhaus (2006) criticizes the Kyoto Protocol for not being su¢ ciently inclusive, cost
e¤ective, or ambitious, the current paper demonstrates that, even without these weak-
nesses, this type of emission agreement is fundamentally ‡awed. The literature usually
emphasizes the positive e¤ects of regulation on technological change.35 With technolog-
ical spillovers, it has been recognized that as a second-best, the climate policy should
33
Other types of heterogeneity would be harder to analyze. For example, suppose the cost of developing
technology, K, varied across countries. In equilibrium, only countries with a small K would invest. This
would also be optimal, but, just as before, the investing countries would invest too little. In a long-term
agreement, one could encourage these countries to invest more by reducing gilt or, if renegotiation is
possible, gide . Such small gi s would not be necessary (or optimal) for noninvesting countries. Naturally,
the investing countries would require some compensation to accept the optimal emission targets. At the
same time, a small gi would not motivate i to invest if i were allowed to purchase permits from noninvesting
countries with higher gj s. Thus, with heterogeneity in investment costs, it matters a great deal whether
side transfers are possible and permits tradable: Proposition 11 would be false if such heterogeneity were
introduced. Evaluating political instruments under heterogeneity is thus an important task for future
research.
34
See Kolstad and Toman (2005) on climate policy and Barrett (2005) on environmental agreements.
Aldy et al. (2003) and Aldy and Stavins (2007; 2009) discuss alternative climate agreement designs.
35
See, e.g., Ja¤e et al. (2003) or Newell et al. (2006).

33
be more ambitious to encourage investments (Golombek and Hoel, 2005; Hart, 2007;
Greaker and Pade, 2009). With regards to the term of the agreement, a typical recom-
mendation is a decade-long agreement, partly to ensure ‡exibility (see, for example, Karp
and Zhao, 2009). The present note, in contrast, shows that short-term agreements reduce
the incentive to invest in new technology and can be worse than business as usual. This
result builds on Buchholtz and Konrad (1994), who …rst noted that R&D might decrease
prior to negotiations.36 Beccherle and Tirole (2011) have recently analyzed a related one-
period model and shown that anticipating negotiations can also have adverse e¤ects if
the countries, instead of investing, sell permits on the forward market, allow banking,
or set production standards. Helm and Schmidt (2014) have endogenized the coalition
size, while Schmidt and Strausz (2014) focus on when cooperation is feasible without side
payments if countries are heterogeneous. With only one or two period, however, all these
models miss important dynamic e¤ects and thus the consequences for agreement design.
By allowing agreements on quantities or emissions, but not on e¤ort or investments,
we follow the literature on incomplete contracts (surveyed by Segal and Whinston, 2013).
With long-term commitments one may point to "the di¢ culty of foreseeing all possible
later contingencies that might arise" (Maskin and Tirole, 1988:550). With such uncer-
tainty, Harris and Holmstrom (1987) discuss the optimal length when contracts are costly
to rewrite. To preserve the optimal incentives to invest, Guriev and Kvasov (2005) argue
that the agents should continuously renegotiate the length. Ellman (2006) studies the
optimal probability for continuing the contract and …nds that this probability should be
larger if speci…c investments are important. This is somewhat related to the result that
the optimal term should be long if intellectual property rights are weak.
A third related literature is that on dynamic games with stocks. Since the evolving
stock typically in‡uences the incentive to contribute, the natural equilibrium concept is
Markov perfect equilibrium and it is quite standard to assume linear-quadratic functional
36
Analogously, Gatsios and Karp (1992) show how …rms may invest suboptimally prior to merger
negotiations. For climate policy, Hoel and de Zeeuw (2010) show that R&D can decrease if countries
cooperate because they then reduce pollution even without new technology; note, however, that there is no
negotiation in their model and their analysis hinges on a "breakthrough technology" and binary abatement
levels. In contrast, Muuls (2009) …nds that investments increase when negotiations are anticipated.
Açikgöz and Benchekroun (2014) discuss similar e¤ects regarding abatements prior to an agreement.

34
forms.37 Calvo and Rubio (2013) provide a nice survey of the applications to environmen-
tal agreements. The closest papers to these notes are probably Dutta and Radner (2004,
2009, and 2012). In these papers, the environmental harm from pollution is linear in the
stock, so the pollution stock does not play the same strategic role as in the present model.
Furthermore, Dutta and Radner mostly focus on equilibrium selection and do not study
incomplete contracts, which is the focus here.

Further Reading and References

Aghion, Philippe; Dewatripont, Mathias and Rey, Patrick (1994): “Renegotiation Design
with Unveri…able Information,”Econometrica 62: 257-82.
Aldy, Joseph; Barrett, Scott and Stavins, Robert (2003): "Thirteen Plus One: A Com-
parison of Global Climate Policy Architectures," Climate Policy 3 (4): 373-97.
Aldy, Joseph and Stavins, Robert (Ed.) (2007): Architectures for Agreement: Addressing
Global Climate Change in the Post-Kyoto World. Cambridge U. Press.
Barrett, Scott (2005): “The Theory of International Environmental Agreements,”Hand-
book of Environmental Economics 3, edited by K.-G. Mäler and J.R. Vincent.
Barrett, Scott and Stavins, Robert (2003): "Increasing Participation and Compliance
in International Climate Change Agreements," International Environmental Agree-
ments: Politics, Law and Economics 3: 349-376.
Başar, Tamer and Olsder, Geert Jan (1999): Dynamic Noncooperative Game Theory.
Siam, Philadelphia.
Battaglini, Marco, and Coate, Stephen (2007): "Ine¢ ciency in Legislative Policymaking:
A Dynamic Analysis," American Economic Review 97 (1): 118-49.
Beccherle, Julien and Tirole, Jean (2010): "Regional Initiatives and the Cost of Delaying
Binding Climate Change Agreements," mimeo, Toulouse School of Economics.
Buchholz, Wolfgang and Konrad, Kai (1994): "Global Environmental Problems and the
Strategic Choice of Technology," Journal of Economics 60 (3): 299-321.
Che, Yeon-Koo and Hausch, Donald B. (1999): "Cooperative Investments and the Value
37
For a comprehensive overview, see Engwerda (2005).

35
 of Contracting," American Economic Review 89 (1): 125-46.
Che, Yeon-Koo and Sakovics, Jozsef (2004): "A Dynamic Theory of Holdup," Economet-
rica 72 (4): 1063-1103.
Chung, Tai-Yeong (1991): "Incomplete Contracts, Speci…c Investment, and Risk Sharing,"
Review of Economic Studies 58 (5): 1031-42.
Coe, David T. and Helpman, Elhanan (1995): "International R&D spillovers," European
Economic Review 39 (5): 859-87.
d’Aspremont, Claude and Jacquemin, Alexis (1988): "Cooperative and Noncooperative
R&D in Duopoly with Spillovers," American Economic Review 78 (5): 1133-37.
Dockner, Engelbert J.; Jørgensen, Ste¤en; Van Long, Ngo and Sorger, Gerhard (2000):
Di¤erential Games in Economics and Management Science, Cambridge U. Press.
Dockner, Engelbert J. and Long, Ngo Van (1993): "International Pollution Control: Co-
operative versus Noncooperative Strategies," Journal of Environmental Economics
and Management 24: 13-29.
Dockner, Engelbert J. and Sorger, Gerhard (1996): "Existence and Properties of Equilib-
ria for a Dynamic Game on Productive Assets," Journal of Economic Theory 71:
209-27.
Doraszelski, Ulrich and Pakes, Ariel (2007): "A Framework for Applied Dynamic Analysis
in IO," Handbook of Industrial Organization 3, North-Holland, Amsterdam: 1887-
1966.
Dutta, Prajit K. and Radner, Roy (2004): "Self-enforcing climate-change treaties," Proc.
Nat. Acad. Sci. U.S., 101, 4746-51.
Dutta, Prajit K. and Radner, Roy (2009): "A Strategic Analysis of Global Warming:
Theory and Some Numbers," Journal of Economic Behavior & Organization 71 (2):
187-209.
Edlin, Aaron S. and Reichelstein, Stefan (1996): "Hold-ups, Standard Breach Remedies,
and Optimal Investment," American Economic Review 86 (3): 478-501.
Ellman, Matthew (2006): "The Optimal Length of Contracts with Application to Out-
sourcing," mimeo, UPF.
Engwerda, Jacob C. (2005): LQ Dynamic Optimization and Di¤erential Games. Wiley.

36
Evans, Robert (2008): "Simple E¢ cient Contracts in Complex Environments," Econo-
metrica 76 (3): 459-91.
Fershtman, Chaim and Nitzan, Shmuel (1991): "Dynamic voluntary provision of public
goods," European Economic Review 35 (5): 1057-67.
Fershtman, Chaim and Pakes, Ariel. (2000): "A Dynamic Oligopoly with Collusion and
Price Wars," RAND Journal of Economics 31 (2): 207-36.
Fudenberg, Drew and Tirole, Jean (1990): "Moral Hazard and Renegotiation in Agency
Contracts," Econometrica 58 (6): 1279-1319.
Fudenberg, Drew and Tirole, Jean (1991): Game Theory. MIT Press.
Gatsios, Konstantine and Karp, Larry (1992): "How Anti-Merger Laws can Reduce In-
vestment, Help Producers, and Harm Consumers," Journal of Industrial Economics
40 (3): 339-48.
Golombek, Rolf and Hoel, Michael (2005): "Climate Policy under Technology Spillovers,"
Environmental and Resource Economics 31 (2): 201-27.
Guriev, Sergei and Kvasov, Dmitriy (2005): "Contracting on Time," American Eco-
nomic Review 95 (5): 1269-1385.
Harris, Milton and Holmstrom, Bengt (1987): "On The Duration of Agreements," Inter-
national Economic Review 28 (2): 389-406.
Harstad, Bård (2012): "Climate Contracts: A Game of Emissions, Investments, Negot-
iations, and Renegotiations," Review of Economic Studies.
Harstad, Bård (2015): "The Dynamics of Climate Agreements," Journal of the Eur-
opean Economic Association.
Hart, Oliver D. and John Moore (1988). "Incomplete Contracts and Renegotiation,"
Econometrica 56: 755-85.
Hoel, Michael (1993): "Intertemporal properties of an international carbon tax," Re-
source and Energy Economics 15 (1): 51-70.
Hoel, Michael and de Zeeuw, Aart (2009): "Can a Focus on Breakthrough Technologies
Improve the Performance of International Environmental Agreements?" NBER Work-
ing Paper 15043.
Houba, Harold; Sneek, Koos and Vardy, Felix (2000): "Can negotiations prevent …sh

37
 wars?" Journal of Economic Dynamics and Control 24 (8): 1265-80.
Ja¤e, Adam B.; Newell, Richard G. and Stavins, Robert N. (2003): "Technological Change
and the Environment," Handbook of Environmental Economics 1 (edited by K.-G.
Mäler and J.R. Vincent), Elsevier.
Karp, Larry S. and Zhao, Jinhua (2009): "A Proposal for the Design of the Successor to
the Kyoto Protocol," mimeo, UC Berkeley.
Kolstad, Charles D. and Toman, Michael (2005): "The Economics of Climate Policy,"
Handbook of Environmental Economics 3: 1562-93.
Levhari, David and Mirman, Leonard J. (1980): "The Great Fish War: An Example
Using a Dynamic Cournot-Nash Solution," Bell Journal of Economics 11 (1): 322-34.
Marx, Leslie M. and Matthews, Steven A. (2000): "Dynamic Voluntary Contribution to
a Public Project," Review of Economic Studies 67: 327-58.
Maskin, Eric and Tirole, Jean (2001): "Markov Perfect Equilibrium: I. Observable Ac-
tions," Journal of Economic Theory 100 (2): 191-219.
Muuls, Mirabelle (2009): "The e¤ect of investment on bargaining positions. Over-
investment in the case of international agreements on climate change," mimeo, Impe-
rial College London.
Newell, Richard G.; Ja¤e, Adam B. and Stavins, Robert N. (2006): "The E¤ects of
Economic and Policy Incentives on Carbon Mitigation Technologies," Energy Eco-
nomics 28: 563-78.
Ploeg, Frederick van der and de Zeeuw, Aart (1991): "A di¤erential game of international
pollution control," Systems & Control Letters 17 (6): 409-14.
Ploeg, Frederick van der and de Zeeuw, Art (1992): "International aspects of pollution
control," Environmental and Resource Economics 2 (2): 117-39.
Reynolds, Stanley S. (1987): "Capacity Investment, Preemption and Commitment in an
In…nite Horizon Model," International Economic Review 28 (1): 69-88.
Rubio, Santiago J. and Ulph, Alistar (2007): "An in…nite-horizon model of dynamic
membership of international environmental agreements," Journal of Environmental
Economics and Management 54 (3): 296-310.
Segal, Ilya and Whinston, Michael D. (2002): "The Mirrless Approach to Mechanism

38
 Design With Renegotiation," Econometrica 70 (1): 1-45.
Sorger, Gerhard (1998): "Markov-Perfect Nash Equilibria in a Class of Resource Games,"
Economic Theory 11: 79-100.
Sorger, Gerhard (2006): "Recursive Nash Bargaining Over a Productive Asset," Journal
of Economic Dynamics and Control 30 (12): 2637-59.
Tsutsui, Shunichi and Mino, Kazuo (1990): "Nonlinear Strategies in Dynamic Duopolistic
Competition with Sticky Prices," Journal of Economic Theory 52: 136-61.
Wirl, Franz (1996): "Dynamic voluntary provision of public goods: Extension to nonlinear
strategies," European Journal of Political Economy 12 (3): 555-60.
Yanase, Akihiko (2006): "Dynamic Voluntary Provision of Public Goods and Optimal
Steady-State Subsidies," Journal of Public Economic Theory 8 (1): 171-9.

39
7. Proofs

All propositions are here proven with the generalizations discussed in Sections 5-6: the
value of yi is given by the increasing and concave function (yi y i ), countries can have
di¤erent reference points y i , and ri generates a direct externality x = s= (n 1) on
j 6= i in addition to the technological spillover e:
X
ui = (yi yi) C (G) kri + x rj :
j6=i

In Sections 3 and 4, B (yi ) (yi y) since y i = y, and x = 0 ) k = K:

While Ui is the continuation value just before the investment stage, let Wi represent the
(interrim) continuation value at (or just before) the emission stage. To shorten equations,
use m @Ui =@G , z e
@Ui =@R , R e
qR R , G qG G + and yei yi + y yi,
where y is the average y i : Note that by substitution,
X X X
G = G e+ yi Ri = qg G + yei R; and
i i i
X X
ui = B (yi yi) C (G) kri + x rj = B (e
yi y) C (G) kri + x rj :

All i’s are identical w.r.t. yei . The game is thus symmetric, no matter di¤erences in Ri
or y i , and the payo¤ relevant states are G and R. Analyzing the symmetric equilibrium
(where symmetric countries invest identical amounts), I drop the subscript for i on U and
W . The proof for the …rst best (3.4)-(3.3) is omitted since it would follow the same lines
as the following proof.

7.1. Proof of Proposition 1

At the emission stage, each country’s …rst-order condition for yi is:

0
0 = (yi yi) C 0 (G) + UG (G; R)
X X
= 0
(e
yi y) C0 G e R+ e
yei + UG (G R+ yei ; R); (7.1)

e and R only. At the invest-

implying that all yei s are identical and implicit functions of G
ment stage, i maximizes:
!
X
e R)
EW (G; kri = EW e+
qG G + ; R Dri kri , (7.2)
i

40
implying that R is going to be a function of G , given implicitly by E@W (qG G +
; R)=@R = k=D and explicitly by, say, R(G ). In the symmetric equilibrium, each
country invests (R(G ) qR R ) =Dn. Thus:
R(G ) qR R
U (G ; R ) = EW (qG G + ; R(G )) (k (n 1) x) )
Dn
@U qR K
z= = (7.3)
@R Dn
in every period. Hence, URG = UGR = 0, m and UG cannot be functions of R and (7.1)
implies that yei , G and thus (e
yi y) C (G) e
(:) are functions of G R only. Hence,
e
write G G R . (7.2) rewritten:

E (qG G + R) kri + U (G (qG G + R) ; R)

and because UR is a constant, maximizing this w.r.t. ri makes qG G R a constant, say

. This gives @ri =@G = qG =Dn and U becomes:

U (G ; R ) = E ( + ) Kr + U (G ( + ) ; R)
qG G qR R
= E ( + ) K + U (G ( + ) ; qG G ))
Dn
qG KqG
m= = @U=@G = K + U R qG = (1 qR ) ; (7.4)
Dn Dn
since G ( + ) and (:) are not functions of G when qG G R = : Since UG is
a constant, (7.1) implies that if R increases, yei increases but G must decrease. Thus,
yi =@R 2 (0; 1), so @gi =@Rj = @ (e
@e yi yi + y Ri ) =@Rj > 0 if i 6= j and < 0 if i = j.
Under (Q), (7.1) becomes
m + cG
0 = cG + by i byi m ) yi = y i : (7.5)
b
X m + cG X
G = (yj e=G
Rj ) + G e+n y Rj )
j
b j

byn e
mn + b G R
G = , so (7.6)
b + cn
0 1
e e
m c @ byn mn + b G R
A = (y i
by m c G R
yi = y i y) + and
b b b + cn b + cn

e P
by m c G j6=i Rj Ri (b + cn c)
gi = y i Ri = (y i y) + :
b + cn b + cn

41
Interrim utility (after investments are sunk) can be written as:

(by)2 m2
Wino c (1 + c=b) G =2 Gmc=b + 2
+ U (G; R): Thus,
2b
b bm (1 + c=b)
@Wino =@R = c (1 + c=b) G + + z: (7.7)
b + cn b + cn
At the investment stage, each country sets k=D =E@Wino =@R. From (7.7):
b+c m (b + c)
k=D = Ec (G) + +z )
b + cn b + cn
k (b + cn) m z b + cn
EG = , combined with (7.6): (7.8)
cD (b + c) c c b+c
b + cn
R = qG G + ny nm=b EG
b
k (b + cn)2 z (b + cn)2 m
= qG G + yn + + ) (7.9)
Dcb (b + c) cb (b + c) c
k (b + cn)2 z (b + cn)2 m
ri nD = qR R + q G G + yn + + :
Dcb (b + c) cb (b + c) c
e = qG G + , (7.6) gives G =EG + b= (b + cn) : Substituting in (7.5) and (7.8):
Since G
m + cG (k=D z) (b + cn) c
yi = y i = yi :
b b (b + c) b + cn
This is helpful when calculating uno
i :
2 2
c k (b + cn) m z (b + cn) b b (k=D z) (b + cn) bc
uno
i = + +
2 Dc (b + c) c c (b + c) b + cn 2 b (b + c) b (b + cn)
!
K e + qG G k (b + cn)2 z (b + cn)2 m
R + yn + + )
Dn Dcb (b + c) cb (b + c) c
2 2
1 k b + cn 1 1 m2 m b + cn k
Euno
i = z + + z
2 D b+c c b 2c c b+c D
!
K e (b + cn)2 k m bc (b + c) 2
qG G R z + yn + :
Dn bc (b + c) D c 2 (b + cn)2

7.2. Proof of Proposition 2

At the emission stage, the countries negotiate the gi s. gi determines yei , and since countries
have symmetric preferences over yei (in the negotiations as well as in the default outcome)
the yei s must be identical in the bargaining outcome and e¢ ciency (7.1) requires:
X X
0= 0
(e
yi y) =n e
C0 G R+ e
yei + UG (G R+ yei ; R): (7.10)

42
The rest of the previous proof continues to hold: R will be a function of G only, so
e
UR = qR K=Dn. This makes EG R a constant and UG = qG (1 qR ) K=Dn, just
as before. The comparative static becomes the same, but the levels of gi , yi , ri , ui and Ui
are obviously di¤erent from the previous case.
The envelope theorem can be used to calculate equilibrium investments:
" #
1 X
max E max yj y) C (G) + U (G; R)
(e kri )
ri n feyj g j
EC 0 (G) D E UG D + E UR D k =

EC 0 (G) D (1 qG ) (1 qR ) K=n (K + xn) (1 1=n) = 0:

Combined with (7.10),

(1 qG ) (1 qR ) K (K + xn) (1 1=n) E 0 (e
yi y)
+ = + UG )
Dn D n
0
(n qR ) K n (n 1) x E (e yi y)
+ = :
D D n
Under (Q), the …rst-order conditions for the optimal yei s:
nm + ncG
0 = ncG + by be
yi nm ) yei = y :
b
X nm + ncG
G = (yj e=G
Rj ) + G e+n y R) (7.11)
j
b

byn e
mn2 + b G R
G = , so (7.12)
b + cn2
by mn e
cn G R
yi = y i y+ and
b + cn2
by mn e
cn G R
gi = y i y+ Ri :
b + cn2
Interrim utility is
2
c 2 b nm + ncG
=Wist G + U (G; R) , so
2 2 b
st
@Wi =@R = cG + m + z:

A country invests until the marginal costs of investment is

k m+z
k = D (EcG + m + z) ) EG = : (7.13)
cD c
43
Subsituting in (7.11), after taking the expection of it, and solving for R gives
m b + cn2 k z
R = qG G + ny + : (7.14)
c b cD c
From (7.12) and (7.13):
k m+z b
G = + ) (7.15)
cD c b + cn2
nm nc k m+z b n k bc
y yei = + + 2
= z+ )
b b cD c b + cn b D b + cn2
c 2 b
ust
i = G (y yei )2 Kr
2 2
2 2
c k m+z b n2 k bc
= + z+ Kr )
2 cD c b + cn2 2b D b + cn2
2
1 k 1 n2 m2 m (k=D z)
Eust
i = z + +
2 D c b 2c c
K m b + cn2 k z 2
bc
qG G qR R + ny + :
nD c b cD c 2 (b + cn2 )

7.3. Proof of Proposition 3

Comparing (7.9) with (7.14) and (7.8) with (7.15),

no st k (b + nc)2 z (b + nc)2 b + cn2 k z

R R = + +
Dbc (b + c) bc (b + c) b cD c
k (n 1)2 qR K
= 1 > 0:
D (b + c) nk
k z b + nc k n 1 qR K Rno Rst
Gno EGst = 1 = 1 = > 0:
cD c b+c D b+c nk n 1
!
2 2
1 k 1 n2 1 1 b + nc k zD b + nc
Eust
i Euno
i = z + + +m 1
2 D c b c b b+c cD b+c
!
K k (b + nc)2 b + cn2 2
bc b+c 1
z +
Dn D bc (b + c) bc 2 (b + cn)2 b + cn2
!
bc2
(k=D z)2 K k
= 2 + z
2 (b + nc) (b + cn2 ) 2bc (b + c) Dnbc (b + c) D
m (k=D z)
(b + c) b + cn2 (b + nc)2 (n 1)
b+c
!
(bc [n 1])2 (k=D z) [n 1]2 k 2K m (k=D z) (n 1)
= z :
2 (b + nc)2 (b + cn2 ) 2 (b + c) D Dn (b + c)

44
Thus, we get U st > U no if

(k=D z) (n 1) k (n 1)2 qR K
Eust
i Euno
i +m z 1 =
(b + c) D (b + c) nk
" #!
2
(bc )2 [n 1]2 [n 1]2 k 2K k 2zk qR K
z z + 1 >0
2 (b + nc)2 (b + cn2 ) 2 (b + c) D Dn D D nk
" #
2 2 2
(bc )2 (b + c) K k qR 2k 2 qR 2 ( qR )2
) > + + 2
(b + nc)2 (b + cn2 ) D K n nK n n2

" #
2 2 2
K K + (n 1) x qR 2 (K + (n 1) x) 2 qR 1
= + 2
D K n nK n n2
2
K nx 2
= (n 1)2 1 + (1 qR )2 :
Dn K

7.4. Proof of Proposition 4

When gi is already negotiated, i invests until

0
k = (gi + Ri y i ) d + zD ) (7.16)
0 1 k zD k zD
yei y = ; Ri = 0 1 + y i gi ;
d d
k zD X
0 1
dri = + y i gi qR Ri; erj : (7.17)
d j6=i

Anticipating this, the utility before investing is:

k zD X
0 1
Ui = EC (G) kri + xrj + U (G; R) :
d j6=i

If the negotiations fail, the default outcome is the noncooperative outcome, giving every-
one the same utility. Since the ri s follow from the gi s in (7.17), everyone understands that
negotiating the gi s is equivalent to negotiating the ri s. Since all countries have identical
preferences w.r.t. the ri s (and their default utility is the same) the ri s are going to be
equal for every i. Symmetry requires that ri , and thus [gi + qR Ri; y i ], is the same
for all countries. (7.17) becomes

0 1 k zD
Dri = :
d

45
E¢ ciency requires (f.o.c. of Ui w.r.t. recognizing gi = qR Ri; + y i and @ri =@ =
1=D8i):

nEC 0 (G) + K=D + n UG nD UR (1=D) = 0 )

EC 0 (G) + m + z = K=Dn: (7.18)

Combined with (7.17), neither G nor R can be functions of R (Ri in (7.17) and (7.18) are
not functions of R ). Thus, UR = qR K=Dn, just as before, and UG cannot be a function
of R (since URG = 0). (7.18) then implies that EG is a constant and, since we must have
= (EG qG G ) =n+qR R =n y, (7.17) gives @ri =@G = (@ri =@gi ) (@gi =@ ) (@ =@G ) =
qG =Dn. Hence, UG = qG K=Dn + UR qG = qG (1 qR ) K=Dn, giving a unique equi-
librium, (7.3) and (7.4), just as before. Substituted in (7.18):

EC 0 (G) = (1 qG ) (1 qR ) K=Dn: (7.19)

This is the same pollution level as in the …rst best (3.3). But investments might be
suboptimally low. Combining (7.19) with (7.16),

0 1 k K qR K D
(gi + Ri y i ) =n EC 0 (G) m= + 1 =
n d D Dn dn
K 1 + (n 1) x=K (D en) (n 1)
K xD=K + e + qR (D=n e)
1 + qR = :
Dn 1 (n 1) e=D Dn en (n Dn 1) D= (n 1) e
0 P
Under (Q), = b (gi + Ri y i ) and EC 0 = c (qG G + i gi ), so
lt
( 0 =n EC 0 ) ( 0 =n EC 0 ) = b gilt + gi =n cn gilt gi )
K=D x=K + e=D + qR (1=n e=D)
Egi gilt = :
b + cn2 1= (n 1) + e=D

7.5. Proof of Proposition 5

At the start of t = 1, countries negotiate emission levels for every period t 2 f1; :::; T g.
The investment level in period T is (7.17) for the same reasons as given above.
Anticipating the equilibrium Ri;T (and Rj;T ) i can invest qR less units in period T for
each invested unit in period T 1. Thus, in period T 1, i invests until:

0
k = d (gi;T 1 + Ri;T 1y i ) + qR k ) (7.20)
X
0 1
Ri;T 1 = qR Ri;T 1 + dri;T 1+ erj;T 1 = y i gi;T 1 (k (1 qR ) =d) :
j6=i

46
The same argument applies to every period T t, t 2 f1; :::T 1g, and the investment
level is given by the analogous equation for each period but T .
In equilibrium, all countries enjoy the same yi y i and default utilities. Thus, just
as before, they will negotiate the gi s such that that they will all face the same cost of
investment in equilibrium. Thus, ri = rj = r and

0 1
Dr = (y i gi qR Ri;t 1 ) (k (1 qR ) =d) :

For every t 2 (1; T ), Ri;t 1 is given by the gi in the previous period:

0 1 0 1
Dr = yi gi qR y i git 1
(k (1 qR ) =d) (k (1 qR ) =d)
0 1
= y i (1 qR ) gi + qR git 1
(1 qR ) (k (1 qR ) =d) : (7.21)

Since ri = rj , (7.20) implies that the equilibrium gi;t + qR Ri;t 1 y i;t is the same (say & t )
for all is:
gi;t + qRt Ri; 1 y i = & t ; t 2 [1; T ]:

All countries have the same preferences over the & t s. Dynamic e¢ ciency requires that
the countries are not better o¤ after a change in the & t s (and thus the gi;t s), given by
( & t ; & t+1 ), such that G is unchanged after two periods, i.e., & t qG = & t+1 , t 2
[1; T 1]. From (7.21), this implies

2
nEC 0 (Gt ) & t + gt K=D + ( & t+1 gt qR ) K=D gt+1 qR K=D 08 & t )
2
EC 0 n + K=D (qG + qR ) K=D + qG dR K=D &t 08 & t )

EC 0 n + (1 qR ) (1 qG ) K=nD = 0:

Using (7.20),

0 k (1 qR ) qG (1 qR ) K
EC 0 (G) n nm = (1 qR ) (1 qG ) K=D
d D
k (1 qR ) k K K x=K + e=D
= (1 qR ) K=D = (1 qR ) = (1 qR ) :
d d D D 1= (n 1) e=D

The gi;T satis…es (7.19) for the same reasons as in the previous proof (and since they do
not in‡uence any Ri;t , t < T ). It is easy to check that UR and UG are the same as before.

47
 0 e+ P 0
Under (Q), = b (y i gi Ri ), EC 0 = c EG gj , and since EcGn n UG =
0 for gi (R), we have
X
b (y i gi Ri ) nc Ge+ gj n UG
h X i k K
b (y i gi Ri ) nc Ge+ gj n UG = (1 qR ) )
d D
k K K x=K + e=D
gi gilt b + cn2 = (1 qR ) = (1 qR ) :
d D D 1= (n 1) e=D

7.6. Proof of Proposition 6

The optimal T balances the cost of under-investment and the cost of not knowing future
s. In period T , countries invest suboptimally not only because of e and x, but because
of the hold-up problem: one more unit of Ri in period T + 1 is not worth much to i, since
the other countries will take advantage of it and pollute more. When all countries invest
less, ui declines. The loss in period T , compared to the earlier periods, is under (Q):

H = (yi;t yi) (yi;T yi) K (ri;t ri;T ) (1 qR )

2 2
b k (1 qR ) b k zD K k (1 qR ) k zD
= + + (1 qR )
2 bd 2 bd D bd bd
2
qR K k 1 k=K d=D qR qR (d=D 1=n)
= 2
1 + :
bD K n d2 =D2 2 2d2 =D2
Note that H increases in e (for given D), x (for given K) and n but decreases in b.
Another cost of the long-term agreement is associated with . Although EC 0 and thus
EGt is the same for all periods,
!2 !2
c c X
t
c c X
t
t t0 t t0
E (Gt )2 = E EGt + t0 qG = (EGt )2 + E t0 qG
2 2 t0 =1
2 2 t0 =1

c c X
t
2(t t0 ) c c 1 2t
qG
= (EGt )2 + 2
qG = EG2t + 2
2
:
2 2 t0 =1
2 2 1 qG
For the T periods, the total present discounted value of this loss is L, given by:
XT
c 1 2t
qG c 2 X T
2 t 1 t 1 2t
L(T ) = 2
= 2
1 qG
t=1
2 1 qG 2 (1 qG ) t=1
T T 2T
c 2 1 2 1 qG
= 2
q G 2
) (7.22)
2 (1 qG ) 1 1 qG
T
c 2 ln 1 2T +2
qG 2
(1 + ln (qG ) = ln )
L0 (T ) = 2 2
:
2 (1 qG ) 1 1 qG

48
 If all future agreements last Tb periods, the optimal T for this agreement is given by
!
X1
Tb0
min L(T ) + T 1 H + T L Tb )
T
=0
0
0 = L (T ) + T
ln H= + L Tb = L0 (T ) + T
ln H= + L Tb
2 3
c 2
1 2T +2
qG (1 + 2
ln (qG ) = ln ) H= + L Tb
= T
ln 4 5
(7.23)
2
2 (1 qG ) 1 1 2
qG Tb0
1

T
assuming some T satis…es (7.23). Since ln > 0 and the bracket-parenthesis in-
creases in T , the loss decreases in T for small T but increases for large T , and there is
a unique T minimizing the loss (even if the loss function is not necessarily globally con-
cave). Since the history (G and R ) does not enter (7.23), T satisfying (7.23) equals Tb,
assuming also Tb is optimal. Substituting Tb = T and (7.22) in (7.23) gives:
T 2T
c 2 qG2
1 qG 2T
2
ln (qG )
H= = 2 2 T
qG 1+ ; (7.24)
2 (1 qG ) (1 qG ) 1 ln

increasing in T . T = 1 is optimal if the left-hand side of (7.24) is larger than the

right-hand side even when T ! 1:

c 2 qG2

2 2
< H= : (7.25)
2 (1 qG ) (1 qG )

If e (for given D), x (for given K) and n are large, but b small, H is large and (7.25) is
2
more likely to hold and if it does not, the T satisfying (7.24) is larger. If c or are large,
(7.25) is less likely to hold and if it does not, (7.24) requires T to decrease.

7.7. Proof of Proposition 7

In the default outcome, a country’s (interrim) utility is:

X
Wide = gide + Ri yi e+
C G gjde + U:

Since i gets 1=n of the renegotiation-surplus, in addition, i’s utility is:

1X X
Wide + Wjre Wjde kri + x rj ; (7.26)
n j j6=i

49
where Wire is the utilities after renegotiation. Maximizing the expectation of this expres-
sion w.r.t. ri gives the f.o.c.

0
k = d gide + Ri y i (1 1=n) + Dz (1 1=n) (7.27)
D X re X1
+E @ Wi =@R e 0 gjde + Rj y j + Dz :
n j6=i
n
P 0
Requiring …rst-best investments, ED (@ ( Wire ) =@R) = K, and since gide + Ri yi
must be the same for all is,

0 0 kn K
k= gide + Ri y i (d D=n) + K=n ) gide + Ri yi = : (7.28)
dn D

Combined with the optimum, (3.2),

0 kn K K
gide + Ri yi E 0 (gi + Ri yi) = (1 qR )
dn D D
K x=K + e=D
= + qR : (7.29)
D 1=n e=D

Since yide y i is the same for every i in equilibrium, the bargaining game (when renego-
tiating the gide s) is symmetric and the renegotiated gire s become e¢ cient (just as under
short-term agreements). Since the …rst best is implemented, UR and UG are as before.
0
Under (Q), gide + Ri yi E 0 (gi + Ri y i ) = b Egi gilt , so

K x=K + e=D
Egi gide = + qR :
bD 1=n e=D

Intuitively, the reason for achieving the …rst best is not that i receives 1=n of the joint
surplus. To see this, suppose i were recognized to make a take-it-or-leave-it o¤er (with
transfers) with probability pi . The expected utility of i could be written, replacing (7.26):
X X
Wide + pi Wjre Wjde kri + x rj :
j j6=i

If pi = 1, this becomes
X X X
Wjre Wjde kri + x rj
j j6=i j6=i
P
and maximizing this w.r.t. ri is not identical to maximizing j Wjre , which would have
given the …rst best. In fact, pi = 1 leads to underinvestment from i since a larger ri

50
improves the other countries’threat point. On the other hand, pi = 0 could motivate i
to invest optimally, since the f.o.c. for the investment would be:

0
k=d gide + Ri y i + Dz;

and equilibrium investments can be arbitrarily large if just gide is su¢ ciently small. Thus,
countries may invest a lot with this renegotiation design not because they share the total
surplus but because their starting point, or threat point, Wide , is very sensitive to ri when
gide is very small.
The intuition can be strengthened by considering an example with n ! 1 and, for
simplicity, x = = 0 ) z = 0. If n ! 1, one can let the externality be positive but
…nite if this is equal to some constant E D d. By de…nition, D d = e (n 1), so
this requires that the externality per country, e = E= (n 1) goes to zero, but the total
externality is still positive. In the limit, i is simply maximizing Wide and the …rst-order
condition w.r.t. ri , (7.27), becomes:

0
k=d gide + Ri yi :

Nevertheless, the …rst-best is implemented by (7.29), which becomes:

0 K E
gide + Ri yi E 0 (gi + Ri yi) = :
D D E
Intuition: In this case, each country internalizes the externality E simply by maximizing
Wide because of the large marginal utility of a better technology in the default outcome,
@Wide =@ri > @Wire =@ri . Countries are here investing as if they end up with the default
outcome, even if they receive a larger quota in equilibrium.
An alternative intuition based on transfers can be stated as follows. In equilibrium,
with the Nash Bargaining Solution, I can calculate the transfer to be ti to country i,
where: !
1 X n 1
ti = Wide Wjde .
n 1 j6=i
n
Thus, if ri changes marginally in a symmetric equilibrium, the transfer changes according
to:
@ti n 1 0
= (d e) de ,
@ri n

51
 0 0
where de gjde + Rj y j 8j in the symmetric equilibrium. In general, the transfer is
not equal to the externality on the other countries. But to motivate …rst-best investments,
these transfers must equal the externality, and this can always be achieved (if d > e) since
0
de is a function of gide .

7.8. Proof of Proposition 8

Take period 1, and assume the countries renegotiate the gi;1 s only (a similar logic holds
if they simultaneously renegotiate future emission levels). If T > 1, Ri;t for t = 2 is
given by gi;2 , nothwitstanding Ri1 and whether the renegotiation over gi;1 fails. Thus, the
equilibrium …rst-order condition w.r.t. ri1 is:

de 0
k = d gi;1 + Ri;1 y i + qR k (1 1=n)
D X re 1 X 0 de
+E @ Wi =@R e gj;1 + Rj;1 yj
n n j6=i
P 0
Requiring …rst-best investments, ED (@ ( Wire ) =@R) = K, and since de
gi;1 + Ri;1 yi
must be the same for all is,

k [1 qR (11=n)] = (d e) 0 gi;1
de
+ Ri y i (1 1=n) + K=n )
0 de k [n qR (n 1)] K
gi;1 + Ri y i = : (7.30)
dn D
Comparing (7.30) and (7.28) reveals that gide is larger in the present case. Under (Q):
K x=K + e=D qR x=Kn
Egi gide = (1 qR ) + :
Db 1=n e=D 1=n e=D
de
Similar argument holds for t > 1, but since R depends on the latest realization of , gi;t
must be renegotiated after is realized in period t 1:

7.9. Proofs of Proposition 9-10

Under short-term agreements (as well as under no agreement), if interrim utility is

e R , investments are given by EWR = k=D while they should optimally be EWR =
W G;
K=Dn, requiring K + x (n 1) = K=n ) x = K=n: Under long-term agreements, the
0
optimal Ri is given by (gi + Ri y i ) D + nzD = K;which is the same as the equilib-
0
rium d + zD = k if x = K ( qR + en (1 qR ) =D) =n: For an agreement lasting T > 1

52
 0
periods, Ri;t , t < T , should be D (gi;t + Ri y i ) + qR K = K; which is the same as
the equilibrium Ri;t if K (1 qR ) =D = k (1 qR ) =d ) x = eK=D. Proposition 10
follows since x > 0 is allowed in the proofs above.

7.10. Proofs of Propositions 11-12

The proof of Proposition 11 is equivalent to the text following it, and thus omitted.
Proposition 12 follows since heterogeneity is allowed in the proofs above.

53