اهداف تولید گازهای گلخانه ای نامشخص بلندمدت، قیمت CO2 و انتقال انرژی جهانی: یک رویکرد تعادل عمومی

The persistent uncertainty about mid-century CO2 emissions targets is likely to affect not only the technological choices that energy-producing firms will make in the future but also their current investment decisions. We illustrate this effect on CO2 price and global energy transition within a MERGE-type general-equilibrium model framework, by considering simple stochastic CO2 policy scenarios. In these scenarios, economic agents know that credible long-run CO2 emissions targets will be set in 2020, with two possible outcomes: either a “hard cap” or a “soft cap”. Each scenario is characterized by the relative probabilities of both possible caps. We derive consistent stochastic trajectories—with two branches after 2020—for prices and quantities of energy commodities and CO2 emissions permits. The impact of uncertain long-run CO2 emissions targets on prices and technological trajectories is discussed. In addition, a simple marginal approach allows us to analyze the Hotelling rule with risk premia observed for certain scenarios.

This paper shows how the current uncertainty about the 2020–2050 CO2 emissions targets may affect CO2 and energy prices as well as technological choices in the energy sector. To assess the cost of reducing GHG emissions, applied general-equilibrium models linking aggregated descriptions of economies and detailed energy sectors together2 have been developed. Some of them, for instance MERGE (Manne et al., 1995), GEMINI (Bernard and Vielle, 2003), IGSM (Sokolov et al., 2005) and WITCH (Bosetti et al., 2006), have been used by IPCC (2007) and USCCSP (2007) to evaluate climate change policies. So far, the issue of agents’ behavior under uncertainty has been addressed in these models through sensitivity analysis (Löschel and Otto, 2009 and Magné et al., 2010), Monte-Carlo simulation (Kypreos, 2006) and stochastic formulations where agents hedge themselves against some probabilistic outcomes. This last approach was first introduced by Manne and Richels (1992) and Manne and Olsen (1996) who studied the effect of a low-probability climate catastrophe on agent's behavior. More recently,3Bosetti and Tavoni (2009) investigate the impact of uncertain energy-related R&D activities and Loulou et al. (2009) derive different EMF 22 radiative forcing scenarios by assuming an uncertain sensitivity of climate to emissions. In this paper, we use a stochastic approach to illustrate how the persistent uncertainty about the 2050 CO2 emissions caps impacts prices and technological choices in the energy sector.4 These energy prices are especially useful to understand agents’ behavior and assess the relevance of our model's results. In a deterministic model, the agents plan their actions with a perfect knowledge of the future, and the efficient (or clean) technologies expand at the optimal rate in the economy. In our model, until 2020, the agents have to invest before knowing the full sequence of emissions caps imposed to regional economies, by trading off the gain in postponing the adoption of efficient but expensive technologies against the risk of being tied to some detrimental technological choice once the actual emissions caps are set. The model we use is a modified stochastic version of the MERGE model.5 For the sake of illustration, here uncertainty only involves two political outcomes, with, at the end of 2020, the setting of either a “hard-cap” policy or a “soft-cap” policy for energy-related CO2 emissions. Each policy defines series of regional quotas which are linearly decreasing until 2050 and constant after this date. Until 2050 the hard-cap and soft-cap quotas are respectively consistent with the IPCC (2007)'s 450 and 550 ppm atmospheric-GHG-concentration scenarios. However, over the model's whole horizon, the hard-cap and soft-cap policies are less stringent than the two IPCC's scenarios since complying with these scenarios would involve post-2050 emissions reductions (IPCC, 2007 and IEA, 2008b). In our model, all agents (i.e., firms and households) are forward looking, in the sense that firms (households) always act so as to maximize their expected present value (expected sum of discounted utilities) under rational expectations. In other words, in each date firms base their current decisions on consistent subsequent prices of inputs and outputs (or, in the case of decisions made until 2020, consistent subsequent prices in each possible outcome), i.e., prices that precisely result from the decisions currently made. Firstly, our approach makes possible an explicit modelling of agents behavior in the presence of long-run CO2 policy uncertainty. Secondly, it yields stochastic scenarios of energy prices—for CO2, oil, gas, power—with two possible sequences for post-2020 prices, that are consistent with the stochastic political scenario under consideration. Note that the unique pre-2020 sequence and the two possible post-2020 sequences obtained for the price of a given energy commodity may broadly differ from the two deterministic sequences of prices that would be determined by successively considering each CO2 target as certain from the beginning (i.e., 2005 in our model). In addition, as illustrated later, a stochastic price scenario is not necessarily bounded by the corresponding two deterministic sequences of prices. This shows the interest of a stochastic-scenario-based approach for studying the energy transition when long-run CO2 emissions targets are uncertain. Section 2 presents the stochastic CO2-emissions policy scenarios under consideration and motivates our approach. The structure, calibration and computation of our stochastic general equilibrium model are discussed in Section 3. The simulation results are studied in Section 4, with an emphasis on the impact of uncertainty on prices and technology trajectories. The last section concludes.

Our simulations show that the uncertainty about long-run emissions targets significantly affects the energy transition at both global and regional scales, as well as CO2 and energy prices. A higher probability for the setting of the hard-cap target at the end of period 2020 leads to more abatement, and therefore more banking, until 2020. In brief, in the electric sector, prior to 2020 coal without CCS declines faster, for the benefit of nuclear and gas technologies. As a higher hard-cap probability leads to a higher stock of banked permits at the start of period 2025, fewer emissions reductions are required in the subsequent periods. More specifically, in all branches, there is less energy conservation in the non-electric sector and, in hard-cap branches, a lower penetration of non-electric backstop technologies. As a result, the technologies deployed in 2060 (when banking is no longer possible) are not fully adjusted to the long-run emissions stabilization targets. Thus, in soft-cap branches of stochastic scenarios, the initial anticipation of a probable hard-cap target results in a 2060 price peak higher than that obtained in the deterministic soft-cap scenario. In addition, since pre-2020 CO2 prices are sensitive to the hard-cap probability, they reveal information about agents’ belief on this probability. Moreover, a pre-2020 banking of emissions permits occurs for a hard-cap probability greater or equal to 0.4 (0.2) in the European Union (Pacific OECD region), while in North America banking occurs in all scenarios. In every region where such a banking takes place, the regional CO2 price follows a Hotelling rule with a risk premium between 2020 and 2025. Since the long-run emissions targets have a negligible impact on regional consumptions, this risk premium is very small. Since a pre-2020 banking occurs in all regions when the hard-cap probability is greater than 0.4, the common belief in a single world CO2 price from 2025 on then leads to a convergence of CO2 prices in OECD regions prior to 2020, even if inter-regional trade of emissions permits does not yet exist. For oil and gas, the observation of the Hotelling rule is hindered by constraints imposed on their production (that limit inter-temporal arbitrage in extraction decisions). Our approach is of course subject to a given number of limitations. One of them relates to the information structure considered. In the model, agents’ belief is not assumed to evolve trough time, as information is fully revealed in 2020. Taking into account a more progressive revelation of information on emissions targets would enrich the model and perhaps significantly influence its results.