The Global Climate Game
Roweno J.R.K. Heijmans
Tilburg University
January 31, 2020
Abstract
I study the private provision of public goods with strategic complementarities and preference-uncertainty. The induced game has a unique equilibrium. While the equilibrium may be first-best, there are well-identified cases for which it coincides with the worst possible outcome. I also introduce sequential global games and show that sequentiality expands the range of preferences for which the public good is provided and coordination occurs on the first-best equilibrium. A possible application of the theory is climate change. Predictions of the model square well with observed real-world climate policies. Compared to existing explanations, my theory requires weaker assumptions on incentives and has more predictive power.
1 Introduction
When public goods get under-provided, economists take free-riders to task. The basic idea is that incentives to shirk are strong and cannot be easily overcome. An important – and often implicit – assumption behind this argument is that individual contributions are substitutable. Yet it is no forgone conclusion that all public goods can be so characterized. Numerous economic interactions are subject to complementarities, invoking opposite incentives (Milgrom and Roberts, 1990). In this paper, I study the provision of a public good with strategic complementarities and preference-uncertainty. I prove existence of a unique equilibrium, in which the public good may not be provided. I identify precise conditions for which the first-best outcome is supported. In a sequential
1 extension of the game, opportunities for efficient provision are greatly expanded. Many an application fits my general theory, but perhaps the most pressing is global warming. This, then, provides the terminology I henceforth adopt. Amid growing concerns that climate change is headed for the wrong direction, little measures to prevent it are actually undertaken. The dominant explanation dates back to Barrett (1994), who in his rightly famous analysis points at free-riding. All countries, so the logic goes, stand to benefit were climate change avoided. Yet each benefits even more letting others do the job. Every nation facing the same incentive, climate change becomes a sure thing. Barrett’s (1994) theory is intuitive and fits neatly into the classic understanding of man and state alike as homo-economic free-riders. Yet its fundamental assumptions are hard to square with the growing body of evidence that free-rider incentives aren’t that significant a driver of behavior (Andreoni, 1995; Fischbacher et al., 2001). Besides, to no unimportant extent are there complementarities – incentives to coordinate – in climate policies, begging the questions whether free-rider incentives even exist to begin with (Potters and Suetens, 2009). The time is ripe for a complementary explanation. My innovation is to take proper stock of two uncertainties that complicate climate politics. First, no country on its own can avert global warming. Unilateral policies are but a drop in the ocean. Game theorists would say that climate policy is a coordination game. Because such games are supermodular, characterized by strategic complemen- tarities, they generally have multiple equilibria, inducing strategic uncertainty.1 It is a priori unclear which equilibrium will be selected (Van Huyck et al., 1990). Above and beyond these coordination problems, there is scientific uncertainty as to the exact consequences of global warming (c.f. Hsiang et al., 2017). Since the benefit of cutting emissions derives from damages avoided, this amounts to another potential source of equilibrium indeterminacy. It may sound reasonable that strategic and scientific uncertainty lumped together should render coordination on any outcome close to impossible, the product of two complications sounding like an even larger complication. I prove this intuition wrong. While each in isolation fuzzes optimal strategies indeed, precisely their intersection is tinder for equilibrium selection. However, the equilibrium selected may be inefficient.
1Climate policy could alternatively be interpreted as a tipping (Barrett and Dannenberg, 2017) or, yet different, weakest link (Harrison and Hirshleifer, 1989) game, both of which incorporate what basically amounts to a discretization of the supermodularity property. All interpretations are consistent with the idea of a carbon-budget, see Allen et al. (2009).
2 How does that work? Loosely speaking, the argument is somewhat as follows. Due to the coordinative nature of best-responses, a country wants to adopt stringent policies only if sufficiently many others do. Given that the number of investing countries is not known in advance, this is what I called strategic uncertainty. Now bring in the scientific uncertainty. Being only imperfectly informed about the true consequences of climate change, countries need to second guess each other’s beliefs. This introduces the possibility that a country is willing to take measures only if, say, N countries do it also, but at the same time thinks it unlikely – as a guesstimate – that N countries will actually join in the effort. Its best response is then to do nothing. Lack of common knowledge can in such a way enforce coordination on undesirable equilibria. Mindful of this result, my theory explains the empirical observation that countries know climate change is bad yet fail to act accordingly. The Paris agreement provides a point in case. While embraced nearly universally, we are nowhere near meeting the promises laid out therein. In contrast to the existing literature, my account of this apparent contradiction does not rely on free-rider incentives, and incorporates the possibility of strategic complementarities in climate policy. The combination of coordination games and payoff uncertainty goes by the name of global games among the initiated. These were introduced to the literature by Carlsson and Van Damme (1993), gaining further momentum when Morris and Shin (1998) applied them to currency crises. The global games approach allows me to solve an important puzzle in environmental economics, if not political reality altogether: why do countries not take measures that preempt runaway global warming? Why so even if avoiding climate change is cheaper than having it? What explains the inertia when everyone knows measures ought be taken? I answer these questions without relying on free-rider incentives, while allowing for complementarities in national climate policies. As an extension of standard global games, I define a sequential global game as one where a subset of players chooses its actions before the remaining players do. This splits the original game into two subgames, where the outcome of the first is observed by players in the second. A natural interpretation in terms of environmental agreements would be technologically advanced countries being able to install renewable technologies earlier on than developing countries. These games produce sharp theoretical predictions: sequentiality facilitates coordination on the first-best. A recent branch of the literature that positively reassesses the possibilities for international environmental agreements (or public good games more generally) builds
3 on work by Harstad (2012), Harstad (2016), and Battaglini and Harstad (2016). These papers are united by a focus on incomplete contracts in dynamic games of technological investment and emission reductions. A political economy justification for the focus on incomplete contracts is given by Battaglini and Harstad (2020). Whether driven by the possibility of renegotiation (Harstad, 2012), catalysed through complementary trade policies (Harstad, 2016) – see also Nordhaus (2015) –, or due to the fact that tough, long-term agreements can mitigate both hold-up and free-rider problems (Battaglini and Harstad, 2016), the common prediction in these papers is cautiously positive: countries can sustain cooperation on desirable outcomes, sometimes even the first-best. My paper shares this optimistic prediction in well-identified cases and is in that sense related. I deviate from their approach in several ways. First, I focus on one-dimensional policies, and consider static or sequential games. Second, I explicitly consider the role of uncertainty in environmental agreements. Finally, while the games studied in this branch of the literature tend to have many equilibria – the symmetric Markov perfect equilibrium is unique, but there can be many others –, the equilibrium in my analysis is globally unique. My focus on the dual face of uncertainty also nests this paper in the literature on imperfect knowledge and environmental regulation. It has long been understood that while regulating externalities under complete information is almost a triviality, informational distortions may ravel matters severely (Weitzman, 1974). Pindyck (2007) identifies three core sources of uncertainty: damage uncertainty, policy cost uncertainty, and discount rate uncertainty. The first of these is the main source of risk considered in Weitzman (2009), Weitzman (2014), Golosov et al. (2014), Hsiang et al. (2017), and Cai and Lontzek (2019). The latter was studied extensively by Weitzman (2001) and Gollier (2002). Given the general formulation of my model, each of these three core uncertainties can be interpreted as part of scientific uncertainty re climate damages. To my knowledge, I am the first to study the intersection between these well-understood sources of uncertainty and coordination problems due to strategic complementarities.
2 A Public Good Game With Strategic Comple- mentarities
Generically, a public good is characterized by only a handful of parameters. There are N ≥ 2 agents providing and enjoying the good. There are to each agent benefits from
4 the good provided. And there is the individual cost of contribution. The difference between benefits and costs is the net benefit of contributing. I define a public good to exhibit strategic complementarities if the net benefit of contributing to the good is increasing in the number of agents supplying it, so best-response correspondences are upward-sloping. (If the net benefit is decreasing in the number of contributing agents, I say there are strategic substitutes, at the heart of free-rider incentives). It is important to distinguish between games of strategic complementarities versus substitutes, as they tend to ignite differential behaviors (Potters and Suetens, 2008, 2009). Denote the set of players P = {1, 2, ..., N}. Each agent i ∈ P chooses action xi ∈ {0, 1}, and all choose simultaneously. Action xi = 1 amounts to (privately) P supplying the good, while xi = 0 means no contribution. Define X = i(xi/N) as average contributions. With a slight abuse of notation, when a continuum of agents is R 1 considered, I normalize P = [0, 1] and let X = 0 xidi. In the interpretation, agents are countries, and the public good is climate change avoided. Contributing to this good can be thought of as investment in renewables, not contributing as relying on fossil fuels. Investment costs. Supplying a public good is costly. Let the gross cost of contribution – investment in the interpretation – be c, where 0 ≤ c ≤ 1. It is costly to install green technologies because this involves changes to the existing electricity network, job losses in the oil and coal industries, forsaken rents from natural resource sales, R&D in green energy generation, and so forth. At the same time, renewable technologies are subject to scale economies. On a unit-basis, it is cheaper to build one million windmills than it is to build only a thousand. Similarly, part of the knowledge created by research spills across boarders, so that the more countries undertake R&D, the less each of them needs to spend individually. These scale economies are incorporated into my model by having the net costs of investment equal to c − X. The costs of generating energy from fossil fuels are normalized to zero. Benefits or damages. Agents derive benefits from public goods, e.g. climatic damages avoided. Model-wise, let the damage due to global warming be given by δ(1 − X), with δ a damage-sensitivity parameter. The term (1 − X) incarnates the idea that higher global investment in renewables means lower emissions and milder climate change; or generally that agents like to see more of the public good provided. Some agents may enjoy the public good more than others; there is no reason to assume absolute homogeneity in preferences. For example, it stands to reason that certain countries are harmed more than others. It is possible to make the parameter
5 describing damages country-specific. I omit this more realistic specification in the general exposition of the model so as to deliver my main insight with minimal complication. Individual-specific preferences will be the main focus of Section 3. Payoffs. Finally, the payoff to agent i – country i’s welfare – is simply the benefits due to the public good minus the cost of provision if the agent is providing it. In climate change parlance, it is the sum of economic damages due to environmental degradation and the costs it incurs from investing. Denote this payoff πi(xi; X, δ), where note that R one may define X−i = j6=i xjdj as the total investment by all countries other than i. When no unclarities arise, I will write πi for simplicity of notation. The model can now be summarized as follows:
i ∈ P (1)
xi ∈ {0, 1} (2) Z X = xidi (3)
i∈P
πi = −(c − X)xi − δ(1 − X). (4)
Uncertainty and signals. Agents may be unsure how much precisely they benefit from a public good, or how much it will cost them to provide it. This could be the case, for example, when benefits realize only in the future, so many things might happen meanwhile. This brings me back to the case of climate change, where the most significant damages are expected to occur decades from now. Sea level rise is hard to predict, and meteorological systems are not well understood. Moreover, possibilities for adaptation to extreme weather are still being explored. Thus far, welfare in the model was assumed to decrease with environmental degradation via the sensitivity parameter δ. The idea that there is no general consensus on precisely how high these costs are effectively means that the true damage parameter δ is not observed. Instead, each country i receives a private noisy signal di of δ, where:
di = δ + i, (5) for i is a random variable drawn uniformly from [−ε, ε], with ε > 0 a measure of damage-uncertainty. It is clear that for any two countries i and j, signals di and dj are correlated since both have the same expected value. However, conditional on their
6 mean, signals are drawn i.i.d. It is understood that the process by which signals (and noise) are generated is common knowledge. Games of this type are generally known as global games, introduced by Carlsson and Van Damme (1993). Throughtout, I assume δ ∈ δ, δ, where δ < c − 1 and δ > 0.
Now suppose country i receives signal di. What does i infer? First, that E[δ|di] = di. In words, a country’s best guess for the environmental damage parameter is the signal it receives. Second, the country learns that E[dj|di] = di. Thus, any country i’s best guess about the signal dj received by country j is simply its own signal, di. There is no need dwelling further in such trivialities, with one exception. Receiving signal di, country i learns that with probability 1/2, the true damage parameter δ is smaller than di (and, symmetrically, that with same probability it is larger). Moreover, given signal di, country i knows that with probability 1/2 neighbor j has received a higher (or lower) signal dj. Formally, Pr[di > dj|di] = Pr[di < dj|di] = 1/2. These are all the building blocks necessary for an analysis of the Global Climate Game, to which I now proceed.
2.1 Two Countries
Intuition may be fostered – and necessary game theoretic techniques may be most simply equipped – by discussing a two-country world first, i.e. P = {1, 2}. In this case, the game outlined above can be represented by Figure 1.
Country 2
0 1