Intertemporal Abatement Decisions underAmbiguity Aversion in a Cap and Trade Simon Quemin WP 2017.06 Suggested citation: S. Quemin (2017). Intertemporal Abatement Decisions under Ambiguity Aversion in a Cap and Trade. FAERE Working Paper, 2017.06. ISSN number: 2274-5556 www.faere.fr Intertemporal Abatement Decisions under Ambiguity Aversion in a Cap and Trade∗ Simon Quemin† LEDa-CGEMP, Paris-Dauphine University − PSL Research University Climate Economics Chair, Paris, France April 2017 Abstract We study intertemporal abatement decisions by an ambiguity averse firm covered under a cap and trade. Ambiguity aversion is introduced to account for the prevalence of regulatory un- certainty in existing cap-and-trade schemes. Ambiguity bears on both the future permit price and the firm’s demand for permits. Ambiguity aversion drives equilibrium choices away from intertemporal efficiency and induces two effects: a pes- simistic distortion of beliefs that overemphasises ‘detrimental’ outcomes and a shift in the effective discount factor. Permit allocation is non neutral and the firm’s intertemporal abate- ment decisions do not solely depend on expected future permit prices, but also on its own expected future market position. In particular, pessimism leads the expected net short (resp. long) firm to overabate (resp. underabate) early on relative to in- tertemporal efficiency. We show that there is a general incen- tive for early overabatement and that it is more pronounced under auctioning that under free allocation. Keywords: Emissions trading; Regulatory uncertainty; Per- mit banking; Ambiguity aversion. JEL Classification codes: D81; D92; Q58. ∗The author would like to thank Lo¨ıc Berger, Jean-Marc Bourgeon, Johanna Etner, Christian Gollier, MeglenaJeleva,PhilippeQuirionandoneanonymousreviewerfromtheFrenchAssociationofEnvironmental and Resource Economists (FAERE). Participants from the 39th IAEE Conference (Bergen, Norway), the 3rd FAERE Conference (Bordeaux, France) and Climate Economics Chair seminars are also acknowledged. The author only is responsible for any inaccuracies or deficiencies. †E-mail: [email protected] 1 1 Introduction As compared to standard markets a specificity of markets for pollution permits is that the supply of permits is fixed, exogenous and imposed by a regulatory authority. In other words, the scarcity created by the cap on emissions is of a political nature and pollution permits are not natural commodities with intrinsic value.1 Rather, their value ultimately relates to the credibility of both the regulator in supervising the market and the regulatory scheme itself. Firms’ anticipation and perception of the future regulatory stringency will therefore guide their abatement, compliance and clean technology investment strategies through time. Moreover, carbon markets2 do not function in a vacuum and are influenced by policies out- side of their perimeters. External and uncertain factors such as macroeconomic conditions, the usage of offset credits for compliance and the reach of complementary policies can erode the stringency of the cap. Policy overlap3 can be fortuitous as in the EU Emissions Trading System (ETS). It can also be explicitly built into an open regulatory system such as the Cali- forniaETSwhich, operatinginconjunctionwithasetofcomplimentarymeasures, constitutes a safety net ensuring the attainment of the state-wide target. In any case, this translates into significant uncertainty about baseline emission levels which Borenstein et al. (2015) estimate to be «at least as large as uncertainty about the effect of abatement measures».4 In most permit markets, permits can be banked for future demonstration of compliance. Un- dercertaintythisflexibilityallowsaleastdiscountedcostsolutionandabatementisefficiently spread over time. In particular, absence of arbitrage requires that the permit price grows at the interest rate (Hotelling, 1931; Rubin, 1996).5 In analysing the effects of regulatory interventions under discontinuous compliance, Hasegawa & Salant (2014) find that «Permit markets may be subject to three kinds of uncertainty: (1) uncertainty about the aggregate demand for permits that will be resolved by an information disclosure at a fixed date in the future; (2) aggregate demand shocks in each period; and (3) regulatory uncertainty.» 1A pollution permit (or allowance) is not a property right per se because it can be limited, modified or simply cancelled. Having inherently ‘ill-defined property rights’ leaves flexibility to the regulator to react to realised shocks, whatever their nature, and adjust the design features of the scheme accordingly. 2Carbonmarketsare our leading example. The terms carbon and pollutionwill be used interchangeably. Notice that our analysis of the impacts of regulatory risks on firms’ market behaviours also applies to other cases of ‘insecure’ natural resource tradable property rights, see e.g. Grainger & Costello (2014). 3Policyoverlapisnotlimitedtoclimatechangeandenergypolicies. Forinstance,Schmalensee&Stavins (2013) underline the impact of railway deregulation on the US SO trading programme. 2 4Borenstein et al. (2015) show that significant baseline uncertainty coupled with little price elasticity is likely to conduce to either very high or very low price levels with high price volatility. 5Note that permits are commodities whose storage costs are negligible. The cost of carry price should therefore correspond to the spot price grown at the interest rate. 2 Only considering points (1-2) and provided that firms are risk neutral, the same rationale applies in the stochastic market equilibrium where intertemporal efficiency obtains in expec- tations (Schennach, 2000). Point (3), however, is essential in that discretionary regulatory interventions are known to distort intertemporal optimality of agents’ decisions due to their anticipation of future regulatory actions (Kydland & Prescott, 1977; Salant & Henderson, 1978). In practice, such regulatory risks, be they upside or downside in nature, have shown to bear on permit prices.6 In particular, Salant (2016) shows that regulatory uncertainty does weigh on price formation in the EUETS even under the assumption that agents are risk neutral.7 This suggests that regulatory risks cannot be entirely hedged against. Note that Koch et al. (2016) find that the EUETS is highly responsive to political events and announcements, which gives empirical support to the theory developed by Salant. Through permit banking, prices reflect expectations about future market developments and may comprise premiums associated with holding permits.8 For instance, Bredin & Parsons (2016) show that the EUETS changed from initial backwardation to contango in late 2008. That is, in Phase II of the EUETS, futures prices were higher than cost of carry prices with implied premiums of significant sizes. This may suggest that firms were hedging themselves against increasing permit prices. Interestingly, Bredin & Parsons (2016) also note that this term structure reflects a sort of fear that is not consistent with the types of reforms discussed at the EU level.9 Moreover, according to Hintermann et al. (2016), positive permit prices in oversuppliedPhaseIIindicatebankingonthepartoffirmsastheyexpectabindingemissions constraint in the future and «because of their awareness of regulatory uncertainty». Faced with regulatory uncertainty, firms lack confidence and/or relevant information to prop- erly assign one probability measure uniquely describing the stochastic nature of their decision problems. Thiscorrespondstoasituationcharacterisedbyambiguity. Bycontrast, riskrefers to situations where such probabilities are perfectly known and unique. The paper thus ex- amines intertemporal abatement decisions by a risk neutral ambiguity averse ETS-liable firm 6Examples are many. The price rise in early 2016 in the New Zealand ETS is attributable to the announcement that the 2:1 compliance rule should be abolished. Similarly, downward pressure on prices in Chinese pilots results from regulatory uncertainty about the transition to a national market, especially regardingthecarry-overprovisionforpilotpermitsintothenationalmarket. PricesinRGGIincreasedwhen the 45% slash in the cap was under discussion, but before it was actually passed and implemented. 7Salant (2016) draws from his analysis of the «peso problem» and the gold spot price in the 70’s that conflicted with the assumption of rational expectations under risk neutrality (Salant & Henderson, 1978). 8Permit prices may thus not be ‘right’ in that they may not reflect (intertemporal) marginal abatement costs. This wedge can also be sustained by other factors such as transaction costs or market power; see Hintermann et al. (2016) for a review of the empirical literature on the price determinants in the EUETS. 9As discussed, changes in the permit supply via the Market Stability Reserve or an adjustment of the annual cap-decreasing factor should shift the term structure as a whole, not just its slope. 3 to take account of the prevalence of regulatory uncertainty. Ambiguity neutrality constitutes our natural benchmark and corresponds to the situation where the firm’s optimal abatement stream is determined by the least discounted expected cost solution (Schennach, 2000). We consider a firm covered under a two-date cap and trade, i.e. the scheme starts at the beginning of date 1 and terminates at the end of date 2. We assume that the firm is already compliant at date 1 but can still undertake additional abatement and bank permits into date 2 in anticipation of date-2 requirements. At date 1, however, both the date-2 market price and the firm’s demand for permits are ex-ante ambiguous and exogenous to the firm. This reflects that regulatory uncertainty (i) directly bears on price formation (Salant, 2016); (ii) also affects the firm’s baseline level of emissions via direct or indirect policy overlaps (Borenstein et al., 2015). We note that regulatory uncertainty could also affect the firm’s allocation of permits. However we choose to keep permit allocation as a parameter in the model to be able to measure its influence on intertemporal abatement decisions.10 Indeed, we will show that neutrality of permit allocation does not hold under ambiguity aversion.11 Ambiguity entirely and exogenously resolves between the two dates.12 We solve the firm’s intertemporal cost minimisation programme by backward induction and compare the opti- mal level of date-1 abatement under ambiguity aversion relative to ambiguity neutrality. We consider a smooth ambiguity model of choice (Klibanoff et al., 2005) in which the firm is con- fronted with different possible scenarios about the future regulatory framework, i.e. objective probabilitydistributionsfortherelatedpermitpriceanddemandforecasts,andhassubjective beliefs over this set of scenarios.13 Attitudes towards ambiguity originate in the relaxation of linearity between objective and subjective lotteries. Ambiguity aversion corresponds to the additional aversion (w.r.t. risk aversion) to being unsure about the probabilities of outcomes and conduces the firm to favour abatement streams that reduce the level of ambiguity. We show that ambiguity aversion drives equilibrium abatement stream choices away from intertemporal efficiency. Before analysing the effects of joint permit price and firm’s baseline ambiguities we consider each source of ambiguity in isolation. This will also allow us to separate the two ambiguity aversion induced effects. First, with pure baseline ambiguity 10Ultimately the firm’s gross effort of abatement (baseline minus allocation) would be impacted by regu- latory uncertainty, which we already capture by letting the baseline be ambiguous. 11Neutralityofallocationdoesnotholdassoonasoneoftheassumptionssustainingthemarketequilibrium solutionofMontgomery(1972)andKrupnicketal.(1983)isrelaxed. SeeHahn&Stavins(2011)forareview. 12Learning is perfect and exogenous to the firm because it can readily observe the prevailing market price and its own demand at date 2, and cannot influence the extent of learning by its date-1 actions. 13Considerforinstancethattheseobjectivescenariosareprovidedbygroupsofexperts,e.g.BNEF,Energy Aspects, ICIS-Tschach, Point Carbon, diverse academic fora or think tanks, etc. 4 and from the perspective of the risk neutral firm, the cap and trade can be assimilated to a tax regime where the tax is set at the expected permit price. Ambiguity aversion induces an upward (resp. downward) shift in the firm’s discount factor when it exhibits Decreasing (resp. Increasing) Absolute Ambiguity Aversion. Early overabatement therefore occurs relative to the benchmark under DAAA, which we define as ambiguity prudence as in Berger (2014) and Gierlinger & Gollier (2017). We also note that the DAAA-induced increase in the discount factor can create a downward pressure on future permit prices. Second, under pure price ambiguity, ambiguity aversion induces another effect by which the firm pessimistically distorts its subjective beliefs and overweights ‘detrimental’ scenarios. When the firm expects to be net short (resp. long) it will overemphasise scenarios where high (resp. low) permit prices are relatively more likely. This raises (resp. lowers) the firm’s estimateofthefuturepricerelativetothebenchmarkandraises(resp.lowers)itsincentivefor early abatement accordingly. As compared to the benchmark the ambiguity averse firm does not solely base its present abatement decisions on the expected future permit price but also on its expected future market position. Note that this ultimately hinges upon the allocation of permits which is thus non-neutral. In particular, we identify allocation thresholds below (resp. above) which pessimism leads the firm to overabate (resp. underabate) early on. Third, under both price and baseline ambiguities, we show that early overabatement occurs when the conditions for early overabatement under pure price ambiguity obtain and, in addition, high-price scenarios coincide with high-baseline scenarios. We then briefly extend the model and consider a continuum of firms identical but for permit allocation where the aggregateambiguityonfirms’baselinesendogenouslydeterminestheambiguouspermitprice. This allows us to refine the threshold condition on permit allocation and we show that there is a general tendency towards early overabatement under a symmetric allocation of permits. This can provide a behavioural explanation for the observed accumulation of unused, banked permitsinallexistingETSsinadditiontootherpermit-oversupplysustainingphysicalfactors (Goulder, 2013; Newell et al., 2013; de Perthuis & Trotignon, 2014; Tvinnereim, 2014).14 The two ambiguity aversion induced effects can be aligned or countervailing, the direction and magnitude of which depend on both the degree of ambiguity aversion and permit allo- cation. An increase in ambiguity aversion always increases the magnitude of the pessimistic distortion in the sense of a monotone likelihood ratio deterioration (Gollier, 2011) and we 14Inparallel,permitpriceshavedeclinedandkeephoveringatlowlevelsorjustabovepricefloorswhensuch price support mechanisms exist. This has sparked short-term regulatory interventions (e.g. ex post supply management) as well as structural design reforms (e.g. in the form of price or quantity-based containment permit reserves) in all existing ETSs that is adding to the level of regulatory uncertainty. 5 show that it can increase that of ambiguity prudence only when ambiguity prudence is not too strong relative to ambiguity aversion. Therefore, a higher degree of ambiguity aversion is not necessarily conducive to a larger adjustment in early abatement (in absolute terms). With a parametrical example we numerically show that early abatement decreases with allo- cation and that the magnitude of the pessimistic distortion is generally greater than that of the shift in the discount factor. This shows that, under ambiguity aversion, early abatement is higher under auctioning than free allocation and suggests that the distortion away from intertemporal efficiency is greater under a cap and trade than an emissions tax. The remainder is organised as follows. Section 2 reviews the related literature. Section 3 presents our model and assumptions. Section 4 analyses the effects of ambiguity aversion on intertemporal abatement decisions relative to ambiguity neutrality. In particular, Section 4.1 considers the case of pure firm-level baseline ambiguity and Section 4.2 that of pure permit price ambiguity. The case of joint price and baseline ambiguities is presented in Section 4.3 while Section 4.4 considers the case of market-wide demand ambiguity with endogenous permit price. Finally, Section 5 illustrates our results numerically and Section 6 concludes. 2 Related literature The paper combines two strands of literature, namely dynamic abatement and investment incentives under environmental policies and decision-making under ambiguity aversion. Dynamic abatement and investment incentives. The paper first extends Baldursson & von der Fehr (2004) to ambiguity aversion. Similarly, Baldursson & von der Fehr show that risk averse firms that expect to be short (resp. long) on the permit market overinvest (resp. underinvest) in abatement technology relative to risk neutrality.15 Note, however, that in our setup firms expecting to be net long (resp. short) can still overabate (resp. underabate) when they exhibit DAAA (resp. IAAA). In practice, there is an asymmetry between long and short entities since the former are under no compulsion to sell and can adopt a passive wait- and-see attitude as long as uncertainty is high and experience is being gained (Ellerman et al., 2010).16 Note also that both cap and trade and an emissions tax deteriorate under 15Ben-David et al. (2000) find similar results which are also supported by laboratory experiments (Betz & Gunnthorsdottir, 2009). Note also that the design of the market (e.g. price containment mechanisms) will affectpermitpriceformationandbankingdecisions(Holt&Shobe,2016). SeeKollenberg&Taschini(2016) for an analysis of the EUETS Market Stability Reserve with risk averse firms. 16InearlyPhaseIoftheEUETSindustrialcompanies(acknowledgedtobelong)didnotseeassignificant an effect of the carbon price on their output cost as power companies did (acknowledged to be short). 6 ambiguity aversion while a tax regime remains intertemporally efficient under risk aversion (Baldursson & von der Fehr, 2004).17 The paper also extends Chevallier et al. (2011) who examine the impacts of a risk on permit allocation on firms’ banking decisions.18 They find that banking increases consecutive to an increase in risk if, and only if the third derivative of the firm’s production function is positive. Relatedly, Colla et al. (2012) show that the presence of speculators with whom risk averse firms can trade permits augments the risk bearing capacity of the market and tends to reduce permit price volatility. The paper follows the literature on dynamic investment incentives under environmental reg- ulations in that it generally considers exogenous shocks on permit prices and firms’ demands, see Requate (2005) for a review. Partial equilibrium models tend to favour tax over cap and trade essentially because in the latter the permit price can comprise a real option value and thus deviates from marginal abatement costs, see e.g. Xepapadeas (2001) with permit price uncertainty and Chao & Wilson (1993) with aggregate demand uncertainty. This lit- erature further distinguishes between irreversible and reversible investments and generally shows that the former tend to decrease with uncertainty (Blyth et al., 2007) while the latter can be used as a hedge and tend to increase with uncertainty (Chen & Tseng, 2011).19 For instance, Zhao (2003) finds that irreversible investment incentives decrease in the level of abatement cost uncertainty, but more so under a tax than an ETS. Note also the ‘partial substitutability’ between abatements and low-carbon investments (Slechten, 2013). Finally, Albrizio & Costa (2014) explicitly analyse the effects of policy uncertainty on irreversible and reversible investments by ETS-liable firms in a model where the regulator’s preferences are observed (and the associated cap set) only once firms have made their investment decisions. Decision-making under ambiguity. FollowingtheseminalcontributionofEllsberg(1961) it is now well documented that most individuals treat ambiguity differently than objective risk, i.e. they prefer gambles with known rather than unknown probabilities.20 There ex- ist alternatives to Subjective Expected Utility (Savage, 1954), see Etner et al. (2012) and Machina & Siniscalchi (2014) for a review. These models of choice differ in their treatments of objective and subjective probabilities and preferences are no longer linear in the proba- bilities. They can roughly be grouped into three categories. The first category represents 17In addition, note that alternatively considering price and quantity regulations allows us to separate the two ambiguity aversion induced effects. 18If firms can pool risks banking may be a risk-management tool besides smoothing abatement over time. 19InthewordsofLaffont&Tirole(1996)low-emissioninvestmentconstitutesa‘bypass’ofpermitmarkets. 20In particular, Ellsberg (1961) showed that rational decision-makers behaved in ways incompatible with the Savagian axiomatisation, e.g. the sure-thing principle. 7 non-additive beliefs, i.e. the probability of an outcome depends on its ranking among all pos- sible outcomes (Schmeidler, 1989; Chateauneuf et al., 2007). The second category considers that agents have a set of multiple subjective priors. Gilboa & Schmeidler (1989) provided be- havioural foundations for Multiple-priors (or Maximin) Expected Utility (MEU) preferences. Ghirardato et al. (2004) later axiomatised the α-maxmin model of choice which considers a convex combination of maximal and minimal expected utilities over the set of multiple priors. ThethirdcategorycorrespondstoRecursiveExpectedUtilitymodels. Inthesemodelsagents have a second-order subjective prior over a set of first-order objective measures and they are EU-maximisersoverthetwolayersofuncertainty(Klibanoffetal.,2005,orKMM).Compared to the other two categories, a KMM model of choice has the advantage of disentangling ambiguity itself (or ‘beliefs’) from attitudes (or ‘tastes’) towards ambiguity. It comes with nice comparative statics and tractability properties to which the decision-making under risk machinery readily applies, can be embedded in a dynamic framework (Klibanoff et al., 2009) and nests other models of choice under ambiguity aversion as special cases.21 Ambiguity aversion has been applied to a variety of fields in economics, such as finance (Gollier, 2011; Gierlinger & Gollier, 2017), formation of precautionary savings (Berger, 2014), self-insurance and self-protection (Alary et al., 2013; Berger, 2016) or health (Treich, 2010; Berger et al., 2013), and can explain otherwise unaccounted for empirical facts such as the equity premium puzzle (Collard et al., 2016) or the negative correlation between asset prices and returns (Ju & Miao, 2012). Closer to our paper is the emerging theory of the competitive firm a` la Sandmo (1971) under ambiguity aversion (Wong, 2015a) and the integration of risk and model uncertainty in Integrated Assessment Models (Millner et al., 2013; Berger et al., 2017). There is also mounting evidence that individuals tend to display ambiguity aversion and especially DAAA, see e.g. Berger & Bosetti (2016) and references therein. The paper develops a two-period model to analyse what is fundamentally a fully-fledged dy- namicproblem. Thisissufficienttocapturetheessenceofthetwoambiguityaversioninduced effects and simplifies the problem at hand in two respects. First, considering more than two periods is technically difficult; for instance, Collard et al. (2016) assume CAAA to simplify Euler equations. This means that Collard et al. solely consider pessimism distortions while shifts in levels are abstracted away. Second, another difficulty relates to the incorporation of new information to update beliefs and preferences.22 This issue is mechanically absent in 21When φ displays CAAA with φ(x) = e−αx, Klibanoff et al. (2005) show that, under some conditions, −α the KMM model approaches the MEU criterion when the ambiguity aversion coefficient α tends to infinity. 22Note that Klibanoff et al. (2009) are able to retain dynamic consistency by defining preferences recur- sively, assuming ‘rectangularity’ of subjective beliefs together with prior-by-prior Bayesian updating, but do 8 our two-period model. Millner et al. (2013) opt for two polar exogenous learning scenarios: one where ambiguity resolves after the first period, the other with persistent and unchanged ambiguity throughout. Guerdjikova & Sciubba (2015) consider two similar types of learn- ing structures, one where the true scenario is determined in the first period, another where the ‘hidden’ scenario is a Markov process and cannot never be identified, as in Ju & Miao (2012). Alternatively, Gierlinger & Gollier (2017) and Traeger (2014) use a one-step-ahead formulation consisting of nested sets of identical ambiguity structures. 3 The model We consider a firm whose production’s by-product is atmospheric pollution. The firm is regulated under a cap-and-trade system. To demonstrate compliance the firm can abate emissions and/or buy pollution permits on the market. There are two dates t = 1,2. At date 1, the date-2 permit price τ and the firm’s date-2 baseline level of emissions b (or production output) are ambiguous in a sense that will be defined below. Ambiguity vanishes at the beginning of date 2, i.e. the firm’s date-2 abatement depends on its date-1 abatement and the price and baseline realisations. We then analyse the firm’s optimal date-1 abatement decisions under ambiguity aversion relative to the ambiguity neutral benchmark. The economic environment. Regulation is effective at both dates and terminates at the end of date 2. As in Chevallier et al. (2011) we assume that date-1 compliance is effective and that all inter-firm permit trading opportunities on the market are exhausted. The firm may still undertake additional date-1 abatement a in the perspective of more stringent 1 date-2 requirements. This frees up a corresponding amount of permits that are banked into date 2.23 This assumption ensures that the Rubin-Schennach banking condition is always satisfied and assumes corner solutions away (Rubin, 1996; Schennach, 2000). There are two alternative descriptions of this framework where regulation is effective at date 2 only. For instance a may also correspond to (i) investments in abatement technology in anticipation 1 of future regulation; or (ii) ‘early reduction permits’ handed out to the firm for its early abatements.24 Given an abatement stream (a ;a ) the firm’s date-2 level of emissions is 1 2 not accommodate the dynamic three-color-urn Ellsberg example in Epstein & Schneider (2003). 23In the case of the EUETS presented in the Introduction, date 1 corresponds to Phase II with a non- binding constraint on emissions and date 2 to Phase III and beyond with an expected permit scarcity. 24These interpretations are equivalent provided that a given level of abatement or investment cuts down emissions by a corresponding amount, and that date-1 abatement or investment reduces both date-1 and date-2 emissions by the same amount. Notice, abatements and investments are substitutes (Slechten, 2013). 9