Elements of emission market design: An experimental analysis of California's market for greenhouse gas allowances

G Model

ARTICLE IN PRESS

JEBO-3346; No. of Pages 19

Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

Contents lists available at ScienceDirect

Journal of Economic Behavior & Organization journal homepage: www.elsevier.com/locate/jebo

Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances夽 William Shobe a,∗ , Charles Holt b , Thaddeus Huetteman c a b c

Center for Economic and Policy Studies, University of Virginia, Box 400206, Charlottesville, VA 22904, USA Department of Economics, University of Virginia, P.O. Box 400182, Charlottesville, VA 22904, USA Power and Energy Analytic Resources Inc., 1650 Wilson Boulevard, Suite 303, Arlington, VA 22209, USA

a r t i c l e

i n f o

Article history: Received 16 October 2013 Received in revised form 29 March 2014 Accepted 1 April 2014 Available online xxx

JEL classification: C92 Q5 Q54 D44 D47 H2 Keywords: Cap and trade Uniform price auction Market liquidity Price containment reserve Emission markets Carbon markets

a b s t r a c t We use a set of economic experiments to test the effects of some novel features of California’s new controls on greenhouse gas emissions. The California cap and trade scheme imposes limits on allowance ownership, uses a tiered price containment reserve sale, and settles allowance auctions based on the lowest accepted bid. We examine the effects of these features on market liquidity, efficiency, and price variability. We find that tight holding limits substantially reduce banking, which, in turn reduces market liquidity. This impairs the ability of traders to smooth prices over time, resulting in lower efficiency and higher price variability. The price containment reserve, while increasing the supply of allowances available to traders, does not appear to mitigate the effects of tight holding limits on market outcomes. As a result, the imposition of holding limits in the allowance market may have the consequence of increasing the likelihood of the market manipulation that they were intended to prevent. Finally, we find that the choice between lowest accepted bid and highest rejected bid for the allowance auction pricing rule does not have a significant effect on market outcomes. © 2014 Elsevier B.V. All rights reserved.

1. Introduction The trading of emission allowances by sources under a regulatory cap on total emissions, or “cap and trade,” is now wellestablished as a regulatory option for addressing a wide range of environmental management challenges. Emission trading is being deployed in a number of countries to control greenhouse gas (GHG) emissions, particularly, emissions of carbon

夽 The authors wish to acknowledge the guidance of Michael Gibbs and his colleagues at the California Air Resources Board. We received helpful comments from Kedin Kilgore, Andy Van Horn, John Melby, Jan Mazurek, Jan Grygier, Dallas Burtraw, Sarah Tulman, Sam Brott, Emily Snow, Flor Guerra, Chloe Gibbs, Christian Vossler, and participants in the University of Tennessee workshop on Identification of Causal Effects in Environmental and Energy Economics. Key financial support for this research was provided by the U.S. Environmental Protection Agency grant number 83456401, by a consortium of California funders lead by Pacific Gas and Electric Company, and by the U.Va. Bankard Fund. Thanks also to Sijia Yang for her tireless and enthusiastic research assistance. ∗ Corresponding author. Tel.: +1 434 982 5376. E-mail address: [email protected] (W. Shobe). http://dx.doi.org/10.1016/j.jebo.2014.04.007 0167-2681/© 2014 Elsevier B.V. All rights reserved.

Please cite this article in press as: Shobe, W., et al., Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.04.007

G Model JEBO-3346; No. of Pages 19

2

ARTICLE IN PRESS W. Shobe et al. / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

dioxide (CO2 ) (Boyd et al., 2006; Ellerman et al., 2014; Schmalensee and Stavins, 2013). Considerable research has been done on the design and performance of emission trading programs and on the importance of particular elements of market design (Aldy and Stavins, 2012; Burtraw et al., 2009), including much laboratory experimentation (Friesen and Gangadharan, 2013). This body of research makes it abundantly clear that the expected and actual performance of emission trading programs depends critically on both the institutional context and on the details of program design. The State of California recently began implementing what arguably will be the most comprehensive set of GHG controls anywhere. Emission trading has a central position in the California program. California regulators made three novel design choices in establishing their emissions market. First, to protect against market manipulation, the program implements a limit on the number of emission allowances that may be held under any single entity’s control, a “holding limit”. Second, to help prevent damaging spikes in allowance prices, the state keeps some allowances in a price containment reserve (PCR) and holds a “discriminatory price sale” of allowances six weeks after each quarterly auction. Third, the quarterly auction uses a “lowest accepted bid” rule for setting the auction closing price, as opposed to the “highest rejected bid” rule used, for example, in allowance auctions held by the Regional Greenhouse Gas Initiative (RGGI). It is the main contribution of this paper to use laboratory experiments to test the effects of these three novel emission market design features. This analysis should be of broader interest, since these design choices may be relevant for a wide variety of tradable assets used to regulate access to environmental resources such as air and water quality, fisheries, and ecosystem services generally. Whenever a regulator establishes a fixed limit on the use of a resource and allocates tradable rights to the resource, then questions of market competitiveness, price variability, and allocation method will need to be addressed by the regulator. Since allowances are an input to production, manipulation of the allowance market might be used to gain competitive advantage in the output market (Montero, 2009). Goeree et al. (2010) describe laboratory experiments where those receiving large grandfathered allocations withheld allowances from the market thereby raising the price of allowances and helping maintain higher prices in the output market. While the details of each instance of market manipulation are different, studies of market manipulation have reached a strong consensus that improving the liquidity of a market is the best defense (Hart, 1977; Jarrow, 1992; Pirrong, 2009; Putnin¸sˇ , 2012). Liquidity is a measure of how many units of an asset (in our case allowances) are available to come into the market as the price rises. Greater liquidity not only makes profiting from price manipulation more difficult and risky, but also improves the overall competitiveness of the market, enhancing price discovery. A recent summary of empirical work finds little convincing evidence for significant manipulation in emission markets (Montero, 2009). It could be that this was a lucky break deriving from over-allocation in markets where banking was allowed, leaving the early cap and trade markets awash in liquidity. As far as we know, there are no experimental tests of asset holding limits on emission market performance. This market design element appears to us to be untried, although it is based on an analogy to position limits in derivatives markets, a policy proposal recently floated by the U.S. Commodities Futures Exchange Commission (CFTC).1 We examine how the use of holding limits may be expected to affect market outcomes. The second design element we investigate is the price containment reserve. Cap and trade programs have, from early days, been subject to the criticism that the vertical supply curve induced by the regulatory cap may induce troublesome price variability in the allowance market, raising compliance costs (Weitzman, 1974). Roberts and Spence (1976) noted that relaxing the slope of the supply curve could limit price volatility while retaining the informational advantages of market exchange. A large recent literature has explored the effects of adding responsiveness to the supply curve with a “price collar”, or a combination of a price floor, below which no allowances are sold by the regulator, and a price ceiling, above which the supply is expanded (Dinan, 2010; Perkis et al., 2014). While there is now a considerable number of experimental studies on emission market design,2 the literature specifically addressing price containment reserves is quite limited. Perkis et al. (2014) investigated the difference between hard and soft limits on price. They showed that price containment reserves are less reliable in controlling excess price variability than are hard caps, implemented as an unlimited additional supply at some high price level. Bodsky et al. (2012) looked specifically at the incentives created by the California 3-tier price containment reserve. Two different models of price containment reserve have been implemented in emission allowance markets to date. The California three tier price containment reserve sale operates something like a discriminatory price auction with only three allowable bid levels. A different model has been chosen for RGGI, where the reserve is released as part of the auction reconciliation process. If the auction would close above a reserve trigger price, then some or all of the allowances in the

1 The legal status of this proposal is, at the time of this writing, unsettled. See: http://www.cftc.gov/LawRegulation/DoddFrankAct/Rulemakings/DF 26 PosLimits/index.htm. Last accessed, February 25, 2014. 2 For recent reviews see Cason, T.N., 2010. What can laboratory experiments teach us about emissions permit market design? Agric. Resour. Econ. Rev., 39(2), 151–161 and Friesen, L., Gangadharan, L., 2013. Environmental markets: what do we learn from the lab? J. Econ. Surv., 27(3), 515–535.

Please cite this article in press as: Shobe, W., et al., Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.04.007

G Model JEBO-3346; No. of Pages 19

ARTICLE IN PRESS W. Shobe et al. / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

3

reserve are made available at the trigger price in order to keep the price from rising above that price.3 Should the reserve be exhausted, the price could continue to rise if necessary to allocate the available allowances to the highest bids.4 At first glance, these two reserve sale mechanisms appear to provide equivalent flexibility of supply; each adds the same level of liquidity at the same trigger price. Although the total supply of allowances is identical in the two cases,5 allowances in the California post-auction sale are available for use in inter-temporal arbitrage, while the RGGI reserve allowances are contingent supply, not available for trade. The California reserve, in effect, allows the limited “borrowing” of allowances from future years’ supply. These price containment sale mechanisms differ in how traders can respond to the risk of future scarcity. A trader who thinks that there is a chance that a given tier of allowances might be exhausted by a certain time in the future, may wish to buy from that tier in advance rather than risk not having any allowances in that tier during a subsequent high demand period. Our study adds to the literature on price containment reserves by comparing the performance of a California-style 3-tier post auction to the in-auction release of the price containment reserve, a design recently chosen by RGGI. This paper also adds to a yet small experimental literature on the choice of the rule for establishing the closing price in uniform price auctions. Cramton et al. (2012) examined the difference between using the lowest accepted bid and the highest rejected bid in an ascending clock, private values auction where two subjects were bidding for one item. They found that the two price setting rules were equally efficient but that the lowest accepted bid raised significantly more revenue for the seller. Interestingly, the highest rejected bid rule led to more “honest” bidding, that is, subjects’ bids tended to be closer to their actual values. The small number of bidders and the single item for sale together provide the best opportunity for observing differences between these pricing rules. Bernard et al. (1998) also consider a comparison of last accepted and first rejected price rules, but in an auction sales setting instead of an auction purchase setting. With 4 or fewer bidders, they observed price differences consistent with the findings of Cramton et al., but differences disappeared with 6 bidders. We test the two pricing rules in an environment more consistent with emission allowance trading, where there is a sealed-bid (single round) auction with numerous bidders and multiple objects for sale. Under these conditions, the difference in bidder incentives as between the two rules should be expected to be even smaller than in the case investigated by Bernard et al. Our results indicate that the holding limits may have the perverse effect of lowering liquidity in the market. If, as the literature suggests, market liquidity reduces the profitability of efforts to manipulate the market, then the holding limits that are part of the California market may actually increase the opportunities for profitable manipulation. We also provide some additional evidence that different price containment mechanisms do induce different patterns of behavior, but at least as implemented thus far, they may not be particularly effective in controlling price variability. Finally, our experiments suggest that, in the setting of a uniform price, sealed bid allowance auction, there is not a significant difference between a pricing rule based on lowest accepted and highest rejected bid. Because elements of emission market design are implemented together and as part of a much larger program, measuring the effects of a given design element will be difficult. A useful case in point arose when the U.S. Environmental Protection Agency implemented the first emission allowance auction. The design of the auction for SO2 allowances was specified in legislation and was apparently chosen for its potential to transfer resource rents from buyers to sellers (Hausker, 1992). An empirical examination of the market design would not even be possible until the auction had been in operation long enough to generate many observations. Experimental investigation quickly explored the potential weaknesses in the mandated auction mechanism, results that were published nearly contemporaneously with the very first SO2 auction (Cason, 1995). In the context of a new law such as AB 32, with its large implications for the state and regional economies, identifying the effects of individual elements of market design may never be possible. The economics laboratory provides an effective alternative approach to predicting the likely effects of these features. A laboratory setting allows us to control how the elements are implemented so that the effects of uncontrolled policy variation do not confound our measurement of the feature of interest. Experimental subjects introduce uncontrollable variation into the laboratory setting, but the influence of this variation on our measurements is minimized by random assignment of subjects to treatment and control groups. By separating the tests of market design features into different experimental treatments and by randomly assigning subjects to treatment and control groups, we can hope to identify the individual influence of each feature on the performance of the emissions market. The next section of this paper reviews the key features of the California GHG program. Section 3 describes our experimental setup, including the hypotheses we will be testing and the measures of market outcomes used. We present our experimental results in Section 4, and Section 5 concludes.

3 There are other differences besides those described here. For example, the California reserve, as initially implemented, is a fixed set of allowances that, once exhausted, never refills. The RGGI reserve refills each year. In our experimental sessions, both treatments use a fixed, one-time reserve. 4 An often discussed variant of the price containment reserve has the top tier of reserve be unlimited, meaning that sufficient allowances would always be made available to prevent the price from rising above that point. This “hard cap” on allowance prices could be built into either the California or RGGI style reserve mechanisms. 5 In our experiments, we use a three price-step trigger for releasing allowances into the auction, so that the supply curves are identical in the two treatments.

Please cite this article in press as: Shobe, W., et al., Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.04.007

G Model JEBO-3346; No. of Pages 19

4

ARTICLE IN PRESS W. Shobe et al. / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

2. California GHG emission regulations under AB 32 According to the U.S. Energy Information Administration (2013), California recently ranked 22nd in CO2 emissions among the world’s political jurisdictions, somewhere between Spain and France. This was true even after its emissions had fallen by more than 10 per cent from those of a decade earlier. So it is no small matter to global GHG emissions that the state of California and the province of Quebec have both established limits on GHG emissions (Newell et al., 2013). The California program, which is often referred to as “AB 32” after its 2006 authorizing legislation,6 will result in 2020 emissions reduced to 1990 levels and 2050 emissions 20% below that. That California emissions per person are already second lowest in the U.S., behind only New York, points to the aggressiveness of this program. California has implemented its GHG emissions limits with a hybrid approach that uses both regulations and a “cap and trade” scheme. Of the anticipated 80 million tons (CO2 equivalent) per year below the business-as-usual forecast, the California Air Resources Board (CARB) forecasts that roughly 60 million tons of reduction will come from regulations and other programs (CARB refers to these as “complementary measures”), meaning that, without the cap, the 2020 goal would be exceeded by about 20 million tons per year. The lion’s share of reductions come from a renewable electricity standard (11.5%), a low carbon fuel standard (15%), energy efficiency initiatives (12%), reducing high warming potential gases (6.5%), and forestry practices (5%). The cap, once fully implemented in 2015, will cover “up-stream” sources including power plants, large industrial facilities, and fuel suppliers, comprising some 350 separate businesses operating more than 600 separate facilities. The cap will cover approximately 85% of GHG emissions from the state, including those emissions imputed to imported sources of energy. The design of the California cap and trade program implements many features used in earlier schemes such as the European Union Emission Trading Scheme (EUETS), RGGI, and the U.S. SO2 allowance trading program under Title IV of the Clean Air Act. The cap declines over time, 2% per year for two years and 3% after that. Allowances representing one ton of carbon dioxide equivalent emissions (CO2 e) must be retired for each ton of emissions. Allowances are uniform except that most allowances have a vintage year, which specifies the first year in which they can be used. A small fraction will not have a vintage, but these are not tradable. The allocation of allowances to emitters reflects a complex set of economic and political considerations. The industrial sector initially receives, at no cost, 90% of expected need based on the average level of emissions per unit of output in that sector, a design chosen with an eye to reducing incentives for production to migrate out of the state. The free share falls modestly over time. Publicly owned electric utilities receive free allowances but must consign for auction any allowances that they do not place in their non-tradable “compliance account”. Investor owned utilities receive their allowances for free but must consign them to the auction and must then purchase any allowances they need, either at auction or from some other seller. CARB rules require that the publicly owned utilities use their free allocations to the benefit of ratepayers. Emitters subject to the regulation, or “covered sources”, must purchase each year at least 30% of the allowances needed to cover their emissions.7 At the end of each three-year “compliance period,” sources must retire sufficient allowances to cover all of their emissions over the three-year period. Sources in deficit after the true-up deadline must retire four allowances for each ton of emissions not covered by allowances by the compliance deadline. CARB holds quarterly, sealed-bid, uniform price auctions to sell any allowances not allocated for free and any allowances consigned for sale at auction. A second auction is held on the same day to sell allowances from the future vintage three years hence. The auctions are open to all. For the first two years, covered sources are subject to the restriction that no single party may purchase more than 15% of the current year vintage allowances at auction and no more than 25% of future vintage allowances, while non-covered purchasers (e.g. brokers) are limited to no more than 4%. Bids are subject to a binding reserve price starting at $10 in 2013 and rising each year at 5% plus the rate of inflation. Allowances not sold due to the reserve not being met are held for sale at auction after two auctions in a row have closed above the reserve price. Once the program is fully implemented, there will be a price containment reserve (PCR) amounting to approximately 7% of the initial cap. These reserve allowances do not have a vintage, so they may be used to cover emissions from any year. CARB has chosen a novel design for releasing allowances from the price containment reserve. Six weeks after each quarterly auction, reserve allowances will be made available in a three-tiered sale. Buyers may post offers to buy allowances at one of the three prices, starting in 2012 at $40, $45, and $50, respectively. Offers at a given price tier will be accepted at that price unless (a) the next lower tier is not exhausted, in which case buyers’ orders will be filled at the lower tier, with “pull-downs” in excess of available allowances randomly allocated, or (b) if the current tier is over-subscribed, then bids in this tier will be allocated in proportion to the total number of bids at that tier. Bodsky et al. (2012) provided early experimental evidence on how these complicated rules affect bidder incentives. An alternative mechanism, now in use by RGGI, releases the reserve directly into the auction a set price points. Later in this paper, we will investigate how this California three tier PCR sale may change incentives and outcomes as compared to the in-auction sale used by RGGI.

6 The Global Warming Solutions Act California, 2006, California Global Warming Solutions Act of 2006 Cal. Health & Safety Code §§ 38,500 et. seq. West Publishing. 7 This feature recognizes that the possibility of bankruptcy may induce firms to under-accumulate allowances. Indeed, the only violations observed in the first compliance period of the RGGI program was from firms not accumulating allowances in anticipation of bankruptcy.

Please cite this article in press as: Shobe, W., et al., Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.04.007

G Model JEBO-3346; No. of Pages 19

ARTICLE IN PRESS W. Shobe et al. / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

5

The AB 32 regulations permit the banking of unused allowances to later periods, but, in a significant break from earlier practice, CARB placed a limit on the number of allowances that a single entity could control at a single point in time. As we will discuss subsequently, the decision to impose limits on the ownership of allowances, referred to as holding limits, was an attempt by CARB to reduce the probability of manipulation of the allowance market.8 The limits amount to approximately 6 million tons for 2013 and 2014, that is in the years before transportation fuels come under the cap, and 11 million tons thereafter. For a sense of the relative magnitude of this limit, Pacific Gas and Electric emissions of CO2 for 2011 amounted to just over 2 million tons. Over a three year compliance horizon, this could limit their allowable ownership to an amount smaller than their compliance obligation. On the other hand, a relatively small emitter would not be likely to ever approach the limit. The holding limit provisions of the California allowance market are unique among emission trading programs and have been the subject of considerable discussion (Linklaters, 2011). Covered emitters can relax the holding limit constraint somewhat by moving allowances that they own from their normal “trading” allowance account to their “compliance” account. Only covered emitters have compliance accounts. Allowances moved into these accounts cannot be traded; they may only be used for compliance at the end of some future compliance period. The exemption a covered source can get by moving allowances into a compliance account is limited by emissions from the source’s most recent verified emission report.9 So, in order for large sources to comply with the holding limit, they will need to move allowances into their compliance accounts to the maximum extent allowed. Even then, since the compliance account exemption is limited by recent past emissions, firms facing a high demand year after a couple of low demand years may have trouble legally accumulating enough allowances to comply with the law. The quarterly auction and PCR sale rules explicitly disallow any bids that would put a bidder into non-compliance with the holding limit (Van Horn, 2013). Municipal electric utilities, although not subject to the same regulations as investor-owned utilities, also face an incentive for shifting allowances into their compliance accounts. They receive their allowances for free, but must choose between placing their free, “grandfathered” allowances in their compliance account or consigning them to the auction. They are prohibited from holding grandfathered allowances in their trading accounts. 3. Method 3.1. Experimental setup Our laboratory sessions were implemented with the University of Virginia’s auction software (Veconlab) and were designed to mimic key features of the California allowance market, maintaining details that would likely affect subject responses to variables of interest: holding limits, price containment reserve, and demand spikes. Key experimental design elements are summarized in Table 1. Each session comprised a sequence of 12 periods (or rounds) of “permit” markets,10 where the permits (allowances) were required for production of an “output” good that sold for a fixed price that was announced at the start of each round. Sessions took about 1.5 h, half an hour for reading the instructions and 1 h for the series of 12 rounds. Each session had 12 subjects, all students at the University of Virginia. Subjects were ‘experienced’ in the sense that they had participated in one earlier training session with somewhat different settings in order to familiarize them with the rather complicated setup. Subjects were paid a US$6 show-up fee plus their earnings during the sessions, with an average total of around $40. All subjects had 5 “capacity units,” each of which could, if operated, produce 1 unit of output per period. All subjects received a free allocation that declined after each compliance period, reflecting the gradual reduction in free allocations in the actual program. Free allocations were zero in the last three rounds. Subjects were assigned one of four different roles, which stayed the same throughout the session. Half were “low users,” requiring one allowance to operate each capacity unit, and half were “high users” requiring two allowances per capacity unit.11 High users received an allocation twice as large as that for low users. Half of the subjects in each of these groups were required to consign all of their free allocations of allowances to auction and half were not required to consign any but had the option to consign. The number of allowances made available in each period declined steadily over time. This schedule was published in advance and available for reference by the subjects throughout the session. The allowances could be banked into future periods in either a holding account or a compliance account, with the compliance account allowances no longer available for any use other than to meet compliance obligations for that subject. “Compliance periods” were defined such that true-up

8 The CARB proposal follows closely the recommendations of a white paper that it commissioned on market manipulation Harris, J.H., 2010. Report on Holdings Limits to the Western Climate Initiative Markets Committee, Alfred Lerner College of Business and Economics. University of Delaware. For reasons that are not made clear, this white paper applies to allowances the ownership limits that the Commodities Futures and Trading Commission (CFTC) applies to derivatives contracts. Harris does not provide any evidence that limiting allowance ownership to less than 1/20th of the amount released in one year by the state is necessary to avoid price-setting behavior by a dominant trader. 9 Van Horn (2013) provides a set of example calculations on how large emitters will need to manage their trading and compliance accounts in order to comply with both the holding limit and their compliance obligation. 10 The language that we used in the laboratory sessions was neutral as to the purpose of the market exercise referring to “permits” rather than “allowances,” and no mention was made of emission markets. 11 One can think of this as reflecting the difference between, say, natural gas and coal-fired generators.

Please cite this article in press as: Shobe, W., et al., Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.04.007

G Model JEBO-3346; No. of Pages 19

ARTICLE IN PRESS W. Shobe et al. / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

6 Table 1 Summary of experiments.

Subjects University of Virginia undergraduates, all having participated in a preliminary training session 12 students in each session, each with 5 production units, 6 high emitters, 2 permits for each unit of output, costs drawn on uniform(4,20) 6 low emitters, 1 permit for each unit of output, costs drawn on uniform(14,22) Half of each group must consign all grandfathered allowances to auction, the other half may consign Session length, approx. 1.5 h with average earning of approx. $40 per subject Sessions 12 periods per session, declining emission cap, allowances bankable for future use Every third period subjects had to show sufficient permits to cover production otherwise incur a penalty of 3 permits for each uncovered unit of emissions Randomly occurring “low hydro” period with higher output price and higher must-run requirements, probability = 0.33 Treatments Holding limit: loose (generally non-binding limit) versus a tight (frequently binding) limit 7 sessions in each treatment Price containment reserve: in auction versus post auction sale (both treatments with three price tiers) 4 sessions in each treatment Auction price rule: highest rejected bid versus lowest accepted bid 6 sessions in each treatment Steps each period: 1. Observe output price and any must-run requirements, along with own current trading and compliance account totals 2. Choose allowances to consign to the auction (in addition to any required amounts) 3. Choose allowances to move from trading account to compliance account 4. Submit bids in auction for up to the current purchase limit and view auction results (including release of permits from the in-auction containment reserve for sessions with such reserves) 5. Post offers to buy and sell in the post-auction spot market and view own spot market results 6. Post offers to buy in the price containment reserve sale (if the reserve was not exhausted) and view results of the sale. [This step only in treatments with a separate price containment reserve sale.] 7. Decide how many capacity units to run, above any must-run production requirement 8. Observe earnings, account totals, cash balances, and any deficits relative to compliance requirements

occurred every third round, at which time any deficits in allowances relative to requirements were tripled. In the last period of a session, the penalty was calculated as three times the closing price of the auction in that period and was subtracted from earnings. Allowances had no redemption value after the end of the last period.12 Table 1 lists the sequence of information disclosures and decisions that comprise a single period in the 12-period sessions. The subjects made a set of decisions about participation in the auction, purchasing allowances in the reserve sale (if applicable), trading in the spot market, and running their production units. At the end of each period, subjects were presented with information about their earnings, their allowance account balances, and their compliance status. The auction in each period had a uniform-price, sealed-bid format. The number of allowances for sale included a base auction quantity, which declined each period, plus any allowances consigned to the auction. The number of bids that a subject could submit was determined by the purchase limit or the holding limit. A bid that would satisfy the purchase limit for a subject was not allowed if winning that bid would result in a violation of that subject’s holding limit. Bidders with winning bids paid only the closing price of the auction for any accepted bids. Half of the sessions used the first rejected bid to set the price, the rule in the RGGI auction, and half used the lowest accepted bid, which is the rule in the California auction. The spot market, which followed immediately after the auction, was a single-round, limit-order call market where each subject could post one buy order and one sell order. A buy order consisted of a quantity requested and a maximum price to be paid, and a sell order was a quantity offered along with a minimum acceptable price. The bids and asks were ordered to determine the market clearing price. All bid and ask orders were filled at that spot market price, and bidders were informed of any changes in their cash and permit holding accounts.13 The costs of operating capacity units were randomly assigned each period and were uniformly distributed in the range of $4–20 per unit for high emitters and $14–22 for low emitters. The value to the subject of a permit for a given capacity unit is equal to the output price minus the cost of operating the unit, all divided by the number of allowances required to operate the unit. For all sessions used in the analysis, there was a two-thirds probability of a $35 output price and a one-third probability of an output price of $60 (or in some cases $50). The higher output price induces a high willingness to pay for allowances and mimics periods of high emissions with accompanying high allowance demand, such as might occur in a drought year with low availability of hydroelectric power.14 During “normal” periods, low users were required to run 3 of their 5 units

12 For this endpoint effect to be important, it would have to have a substantially stronger effect in one treatment than the other. There is nothing about our results that indicates such an asymmetric effect. 13 Ties were determined by a random device. 14 As we discuss shortly, a preliminary set of sessions with a somewhat looser cap were run initially. While these sessions had other differences which limit their use in some hypothesis tests, these sessions are informative, and we note when we use them to support our analysis.

Please cite this article in press as: Shobe, W., et al., Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.04.007

G Model

ARTICLE IN PRESS

JEBO-3346; No. of Pages 19

W. Shobe et al. / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

7

while high users had no “must-run” requirement. During “low hydro” periods, the must-run requirement for low users rose to 4 and high users had a must-run requirement of 1 unit. 3.2. Experimental measures of market performance 3.2.1. Liquidity Market liquidity facilitates price discovery and discourages efforts to manipulate market prices. Market manipulation generally depends on making a purchase or a sale that would likely not be profitable were it not for the effect on market prices. High levels of liquidity, make it riskier and more expensive to move market prices. Our experimental sessions have four sources of liquidity, the number of allowances provided at auction, allowances held in trading accounts, allowances consigned to the auction, and allowances offered in the spot market. The last three of these may only come from allowances held in subject trading accounts. The base auction amount is identical across treatments. We used the number of allowances in trading accounts as our key measure of liquidity. 3.2.2. Price variability Our experimental setup determines the costs of running each unit of capacity, the output price, and the supply of allowances. Since there is no abatement technology, the value of allowances is completely determined by their use in production. The abatement decision is the same as the decision to run or not to run a production unit. (At the market level, abatement can occur by a shift of production from high emitters to low emitters.) The value of using an allowance is the output price minus the cost of production, taking into account the number of allowances needed for each unit (one for “low users” and two for “high users”). There is no interest cost of storing allowances from one period to the next. We can calculate which production unit is the marginal unit over all periods in a session. The value of allowances in production for this unit is the dynamically efficient price of emissions. If one market design element results in market prices that are systematically further away from the marginal abatement cost, then that design element harms the market’s price discovery function. We optimal use the absolute deviation between the market price and the dynamically efficient price, that is |Piactual − Pi | for period i as a measure of market performance. The price can be observed in each period of a 12-period session. For a session, the price discovery measure is the average of the absolute deviations over the 12 periods. 3.2.3. Efficiency As the price moves away from the marginal abatement cost benchmark, businesses receive incorrect signals about which units of abatement are worthwhile and which ones are not. This results in cases where some production units worth running do not run and some units not worth running do run. Both cases result in lost surplus because either costs are higher than they should be or benefits are lower than they should be. Absent the need for allowances, efficiency would be measured by the difference between the cost of production and the output price. But emissions have a social cost, which, in our experiments, is equal to the long-run marginal cost of abatement. This amount must be subtracted off of private surplus of operating the unit that results in the emissions. Since our measure of surplus is the net value to society of a given unit of production accounting for emission damages, then other things equal, allowances taken from the price containment reserve would lower the social surplus if all of the other allowances were applied to the most efficient production units. Allowances charged as penalties and retired would reduce otherwise optimal production away from the efficient equilibrium.15 The actual effect of these marginal allowances on the social surplus depends on the other production decisions during the session and cannot be predicted in advance. 3.3. Treatments and hypotheses 3.3.1. Loose vs. tight holding limits Subjects may choose to bank allowances from one period to the next in response to the expected tightening of the cap and to expected future periods of high allowance prices. With full information, zero interest rates, and perfect foresight, the market price should be close to the fixed, long-run marginal abatement cost. Deviation of the market price from the long-run marginal abatement cost would result in opportunities for profitable trades, bringing the price closer to its dynamically efficient level.16 This perfect foresight dynamic equilibrium would maximize the net social product. In a “myopic equilibrium”, no intertemporal smoothing is possible. With no restrictions on banking, experimental subjects can be expected to smooth prices relative to the myopic case. Tight holding limits limit smoothing opportunities. Each session had either tight holding limits of 7 for low emitters and 14 for high emitters or loose holding limits of 13 and 26 for low and high emitters, respectively. The tight holding limit, at just greater than the auction purchase limit, was still greater than any routine contemporary compliance needs. Loose holding limits allowed subjects to hold more than twice

15 Reserve and penalty allowances only affect the surplus if they change the number of allowances used, which only happens under a limited set of circumstances. 16 Assuming further that the cap was set optimally by policy makers, then the market price is also equal to the social opportunity of the use of the allowance for emitting greenhouse gases, commonly referred to as the “social cost of carbon”.

Please cite this article in press as: Shobe, W., et al., Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.04.007

G Model

ARTICLE IN PRESS

JEBO-3346; No. of Pages 19

W. Shobe et al. / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

8

Table 2 Experimental sessions and session-level results. Session

Holding limit

Reserve sale

Auction price rule

Reserve quant.

Output prices (low, high)

Total session cap

Liquidity

Price variability

Efficiency

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

Loose Loose Tight Tight Tight Loose Loose Tight Loose Tight Loose Tight Loose Tight Tight Tight Tight Tight

Post auction Post auction Post auction Post auction Post auction Post auction Post auction Post auction Post auction Post auction Post auction Post auction Post auction Post auction In auction In auction In auction In auction

Highest rejected Highest rejected Highest rejected Highest rejected Highest rejected Highest rejected Highest rejected Highest rejected Highest rejected Highest rejected Lowest accepted Lowest accepted Lowest accepted Lowest accepted Highest rejected Highest rejected Lowest accepted Lowest accepted

24 24 24 24 24 24 36 36 36 36 36 36 36 36 36 36 36 36

35, 50 35, 50 35, 50 35, 50 35, 50 35, 50 35, 60 35, 60 35, 60 35, 60 35, 60 35, 60 35, 60 35, 60 35, 60 35, 60 35, 60 35, 60

558 558 558 558 558 558 498 498 498 498 498 498 498 498 498 498 498 498

46.8 27.0 45.6 29.9 33.4 48.8 59.1 19.3 51.3 36.0 71.8 28.7 51.4 34.3 38.7 30.8 20.4 33.6

1.08 2.17 2.33 5.75 1.58 1.33 1.71 6.96 1.83 2.08 1.17 2.92 2.54 4.63 3.13 3.75 4.17 4.21

3086 2731 3015 2441 2799 3098 4267 3431 4009 4290 4537 3940 4304 3995 4396 3758 4226 3919

Note: A session comprises 12 trading periods with the number of new allowances made available by sale or grant falling in each subsequent period.

as many permits as needed to cover maximum possible production of 5 during any period. Since we were examining the effect of tight holding limits, the treatment with the loose holding limits was our baseline treatment, and the tight holding limit case was the alternative treatment. For each measure of market outcome, we test the null hypothesis that tight holding limits do not affect market outcomes. In this case, theory strongly suggests that the constraint of tight holding limits cannot improve market performance, which implies the one-sided alternative hypothesis that binding holding limits make market outcomes worse (lower liquidity, lower efficiency, and higher price volatility). 3.3.2. In auction release of PCR vs. post auction sale A one time (non-replenished) allowance reserve was established at the start of each session. The reserve was divided into three price tiers of one third of the available reserve each, and once an allowance was allocated to a price tier, it remained in that tier until sold or until the end of the session. Allowances in these tiers were sold at $23, $28, and $33, respectively. These values were chosen to be in the range of prices expected during high demand periods. The reserve allowances were released in one of two ways, each corresponding to an experimental treatment: a California-style, three-tier sale that takes place after the auction or a RGGI-style release of allowances into the auction itself at the same prices as in the three-tier sale. For the sessions with a post-auction price containment reserve sale, orders were filled from the price tier in which the bid was made unless the “pull-down” rule provided that the order be filled at the next lower price. Oversubscribed tiers were filled in proportion to the bidder’s share of the total number of bids at that tier. All of the sessions used for testing the effects of the price containment reserve treatments on market outcomes were sessions with tight holding limits. We test the null hypothesis of no difference in market outcomes between the in auction release of the reserve compared to a post auction sale. Since there is not a strong theoretical case for one type of reserve or another, the alternative hypothesis is that the outcomes are not equal. 3.3.3. Highest rejected bid vs. lowest accepted bid Each session comprised 12 periods with a permit auction in each period. The baseline treatment used the highest rejected bid to establish the auction closing price, that is, the price paid by all winning bidders. The alternative treatment used the lowest accepted bid, the choice of California in its auction design. In very thin markets, it is possible that the pricing rule could change bids on permits that bidders see as marginal. Recent experiments with ascending price auctions found differences in bidding behavior under these rules (Cramton et al., 2012). We use a two-sided test of the null hypothesis of no effect for the difference between the two auction pricing rules. 3.4. Sessions Our experimental sessions are listed in Table 2. A preliminary round of six sessions (1–6) differs from the later sessions in three key respects, they had a somewhat less binding cap on emissions, a lower output price in low hydro periods, and, unlike the other sessions, the allowances in the price containment reserve were reallocated at the end of each round so that each tier always had one third of the remaining allowance left in the reserve. This tier reallocation results in a different pattern of purchase from the reserve, as we will discuss later. The results from these sessions differ in important ways from the remaining sessions and so are not strictly comparable. We will show how the exclusion of these sessions affects our hypothesis tests. Please cite this article in press as: Shobe, W., et al., Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.04.007

G Model JEBO-3346; No. of Pages 19

ARTICLE IN PRESS W. Shobe et al. / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

9

Fig. 1. Permits made available in each period by session cap.

Sessions 7 through 14 all use the California-style post-auction price containment reserve sale and allow us to test the effects of tight versus loose holding limits. Sessions 8, 10, 12, and 14 through 18 allow a test of the post-auction reserve sale against the RGGI-style in auction allowance release. Sessions 7 through 18 allow a comparison of the highest rejected bid and lowest accepted bid pricing rules. For this comparison, we use the tight holding limit sessions from the second session group along with four additional sessions that differ from the other four sessions only in the reserve sale treatment. Half of these sessions use the highest rejected bid for determining the auction closing price and half use the lowest accepted bid. As shown in Fig. 1, the emission cap declines somewhat more sharply in Sessions 7 through 14 as compared to the first 6 sessions.

4. Results Each of our experimental sessions comprises 12 periods of trading in a single market, with the separate periods linked by banking and by expectations of future scarcity of allowances. This structure allows us to test for differences between the treatments either at the aggregate level of the 12-period market (a session) or at the disaggregated period level. Our first approach to analyzing our results is done at the session level. We use the Wilcoxon rank sum statistic to test for difference between treatments in our three outcome variables of interest, liquidity, price variability, and efficiency. Our second approach is to use ordinary least squares regressions (OLS), treating each period as an observation, but controlling for treatments and other observable differences such as the position of the period in the session, whether the period had a demand shock, and the relatively looser cap of Sessions one through six. We estimate three separate regressions, each with one of our outcome measures as the dependent variable. Each regression equation has independent variables to control for observable differences among the periods. This includes a dummy variable for each treatment, Tight holding limit, Post auction reserve sale, and Lowest accepted bid, along with dummy variables for the period within the session, whether a given period is a ‘low hydro’ period with extra high allowance demand, and whether the session was in the first six sessions where the total session cap is somewhat higher than for the remaining sessions. The random assignment of subjects into treatments allows us to identify the causal effects of treatment using OLS. Although sessions can be treated as statistically independent, given our randomization, econometric concerns over heteroskedasticity and within-session correlation of errors nevertheless remain. To accommodate this, we estimate clusterrobust standard errors, but since tests based on cluster-robust standard errors tend to over-reject the null hypothesis when the number of clusters (i.e. sessions) is small, we adopt the wild bootstrap approach of Cameron et al. (2008). The wild bootstrap estimates the distribution of the regression coefficient t-statistics by resampling the dependent variable with a randomly drawn weight applied to residuals from the original OLS regression with the null hypothesis imposed. Our bootstrap used Rademacher weights of −1 and 1 applied by session. The significance levels reported are based on these estimates. Please cite this article in press as: Shobe, W., et al., Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.04.007

G Model

ARTICLE IN PRESS

JEBO-3346; No. of Pages 19

10

W. Shobe et al. / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

Fig. 2. Trading account totals by period by holding limit.

4.1. Effect of tight holding limits 4.1.1. Liquidity Each laboratory session had either a tight holding limit or a loose holding limit, with tight holding limits being just greater than the auction purchase limit for each subject and loose holding limits being twice the auction purchase limit. The looser holding limit is still tight enough to bind on occasion, but with much lower frequency and at much higher banking levels. Given the constraint that the tight holding limits place on participant ability to bank allowances in tradable accounts, and given the incentive that participants have to move permits into non-traded accounts, we expected that liquidity, as measured by the number of banked allowances available for trade, would fall. The effect of the tighter limit on subjects in our sessions was considerable.17 Fig. 2 shows the difference in the size of the tradable bank for each period separated by treatment. For most of the auctions, the average size of the tradable bank under the loose holding limit is double that under the tighter limit. Fig. 3 shows the frequency distribution of different trading account sizes with tight holding limit observations in the upper panel and loose in the lower panel. Of the banking that is done, trading accounts make up a much larger share of the total bank in the sessions with loose holding limits. Trading accounts are both larger and a larger share of the total bank under loose holding limits, which is illustrated in Fig. 4. This is a scatterplot of the number of allowances in trading accounts by the total number of banked allowances, so it is a way of viewing trading accounts relative to non-tradable compliance accounts and to the bank as a whole. As the size of the total bank rises, we see the number of allowances in trading accounts rises as well, but there is a striking difference in the shares of allowances in trading accounts between the two treatments. The trading accounts are distinctly larger under loose holding limits. Indeed, total banking never reaches 100 allowances under tight holding limits. Even for very small banks, where subjects are unlikely to be constrained by the holding limit, the trading accounts are smaller under tight holding limits and the share of allowances in compliance accounts is higher. The lines represent the line of best fit through the points for each treatment, and the shading shows the 95% confidence interval for that line. This loss of liquidity is not made up for in other ways. Volumes of spot market trades are not significantly different across treatments. Nor is there an increase in the consignment of permits to the auction, as illustrated in Fig. 5. We find a very significant difference between the treatments and easily reject the null hypothesis of no effect. Result 1: Tight holding limits reduce the stock of tradable permits compared to loose holding limits and, hence, lower liquidity. Based on the observations in Sessions 1 through 14 the Wilcoxon [Mann–Whitney] test, stratified by the two groups of sessions with different total session caps (Sessions 1–6 and 7–14), has a p-value of 0.014, so we reject the null hypothesis that tight holding limits have no effect. Regression results are reported in Table 4. The Tight holding limit variable, which is 1 for tight limits and 0 for slack limits, is highly significant and confirms the results from the rank sum tests. We conclude

17 Unless otherwise noted, all results will exclude the preliminary sessions, since they are not completely parallel to the sessions we used for our hypothesis tests.

Please cite this article in press as: Shobe, W., et al., Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.04.007

G Model JEBO-3346; No. of Pages 19

ARTICLE IN PRESS W. Shobe et al. / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

11

Fig. 3. Distribution of trading accounts by holding limit.

that, in our laboratory setting, tight holding limits exert a strong negative influence the level of banking. Fig. 4 also suggests a substantial difference in the proportion of banked permits that are available for trade. Tight holding limits will, when binding, lower liquidity. The dummy for the first six sessions is not significant. The period variable tracks the increasing scarcity of allowances over time, as the cap falls. The response to the declining cap is significant and of the expected sign for all three measures of market performance: liquidity falls, price variability rises, and efficiency rises as subjects liquidate their allowance banks and use the banked allowances in production, generating social surplus. The low hydro variable is also highly significant and behaves as expected. Liquidity falls as producers use their banked allowances to increase production in response to the high output price. Price spikes up during these periods, especially during tight holding limit sessions, which can be seen clearly in Fig. 6. Surplus spikes up in low hydro years as subjects

Fig. 4. Trading accounts by total banked permits.

Please cite this article in press as: Shobe, W., et al., Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.04.007

G Model JEBO-3346; No. of Pages 19

12

ARTICLE IN PRESS W. Shobe et al. / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

Fig. 5. Total consigned permits each period by holding limit.

liquidate allowance assets to produce the higher valued output, and the allowances used have a value in production that is higher than the social cost of the associated emissions. 4.1.2. Price variability In this section, we focus on whether the reduced liquidity is likely to be associated with lower price variability. Given the importance to market participants of being able to accumulate allowances based on their best assessment of current and future scarcity, we would expect that lower liquidity would be associated with prices that stray further away from their surplus (and profit) maximizing value. Fig. 6 effectively summarizes the results from the laboratory sessions. The dotted and dashed lines represent two extreme possibilities. The dashed line with asterisks gives the predicted market-clearing price in each period of a session under the assumption that subjects only take into account the current period cap, costs and output prices. It represents the “perfectly myopic” Walrasian competitive equilibrium, where subjects act as if only the current period matters and do not account for any benefits they would receive by banking allowances from one period to the next. The horizontal dotted horizontal line in Fig. 6 shows the dynamically efficient price ($14.75), which is the value of the marginal abatement cost combining all 12

Fig. 6. Auction price each period averaged across sessions.

Please cite this article in press as: Shobe, W., et al., Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.04.007

G Model

ARTICLE IN PRESS

JEBO-3346; No. of Pages 19

W. Shobe et al. / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

13

Table 3 Wilcoxon rank sum tests.

Holding limit Reserve sale type Auction price rule

Liquidity

Price variability

Efficiency

H0 : p = 0.015** (−2.23) H0 : p = 0.443(−0.289) H0 : p = 0.591 (0.160)

H0 : p = 0.014** (2.13) H0 : p = 1 (0) H0 : p = 0.818(−0.320)

H0 : p = 0.038** (−1.83) H0 : p = 0.443(−0.289) H0 : p = 0.699(−0.480)

One-tailed+ n = 14 Two-tailed n = 8 Two-tailed n = 12

Notes: Hypothesis tests are based on session averages where each session comprises 12 market periods. Test statistic reported in parentheses. + Stratified by session total permit cap. ** Rejection of null hypothesis significant at 5%. Table 4 OLS results for key market outcome measures.

Constant Tight holding limit Post auction reserve sale Lowest accepted bid Period in session Low hydro Sessions 1–6 R2 N

Liquidity

Price variability

Efficiency

68.58*** (6.21) −18.42*** (7.28) 3.91 (5.21) 0.82 (4.83) −2.88*** (0.70) −6.13*** (1.84) 4.49 (5.76) 0.401 216

−1.63** (0.80) 2.06** (0.91) 0.196 (0.80) 0.028 (0.67) 0.368*** (0.12) 1.73*** (0.59) 0.73 (0.97) 0.378 216

45.00*** (8.15) −18.08** (9.12) −8.39 (8.63) 6.02 (6.72) 1.39** (0.53) 33.32*** (9.29) 14.658 (10.44) 0.375 216

Notes: Unit of observation is one period with 12 periods in each of 18 sessions. Standard errors in parentheses. ** Standard errors and significance levels calculated using the clustered wild bootstrap (N = 2000): 5%. *** Standard errors and significance levels calculated using the clustered wild bootstrap (N = 2000): 1%.

periods in the session and represents the “perfect foresight” case, where subjects correctly anticipate all future scarcity and completely balance their allowance scarcity across all periods. Market participants who anticipate future scarcity can profit by accumulating allowances early and saving them for later periods, when they will have greater value as the cap declines. This would cause the observed price in the lab to be smoother and flatter than the predicted myopic clearing price. The more the market induces traders to smooth and flatten the price path away from the myopic price path toward the efficient (flat) price path, the closer the market comes to the least-cost pattern of emissions. The solid lines in Fig. 6 represent the average across sessions of auction prices for each auction period depending on whether the holding limit was loose (triangles) or tight (circles). The difference between these two lines clearly illustrates our next result, that tight holding limits tend to drive the auction price further from the dynamically efficient price path, reducing the ability of traders to prepare for high demand periods, and limiting trader ability to prepare for long-term future scarcity by making additional reductions in the present and banking the unused allowances for use in later years. Price discovery is less effective with tight holding limits than with loose holding limits. Result 2: Tight holding limits increase the variability of actual prices around the dynamically efficient price. As before, we tested the hypothesis using the session averages only from sessions 1 through 14. The Wilcoxon rank sum statistic (reported in Table 3) for the null hypothesis of no effect has a p-value of 0.014 (one-tailed), so we reject the null hypothesis that tight holding limits have no effect on the deviation between the observed price and the dynamically efficient price. The OLS estimate of the effect of tight holding limits on price variability (reported in Table 4) is positive and statistically significant, again consistent with the results of the rank sum test. In considering the effects on price, our focus so far has been on the auction price. This leaves open the possibility that spot market trading serves to move prices back toward the socially efficient level. Fig. 7 shows the average difference between auction and spot prices by holding limit treatment. While there may be some tendency under loose holding limits for the spot market to move price closer to the dynamically efficient price, this is not true during tight holding limit sessions. Fig. 8 shows the standard deviation of auction prices for each of the holding limit treatments broken out these out by auction period.18 Both figures clearly show the increased price variability with tighter holding limits. 4.1.3. Efficiency There is every reason to expect that the difficulty that lab subjects have in shifting their stock of allowances across time periods should end up reflected in lower efficiency for tight holding limits relative to loose holding limits. The permit price signals whether a particular production unit is worth running. Any movement of the realized price away from the efficient price will provide incorrect signals at the margin about running a unit. Fig. 9 shows the average surplus lost due to the running of lower valued production units in each period separated out by loose and tight holding limits.

18

This image includes observations from Sessions 1 through 14.

Please cite this article in press as: Shobe, W., et al., Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.04.007

G Model JEBO-3346; No. of Pages 19

14

ARTICLE IN PRESS W. Shobe et al. / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

Fig. 7. Prices in the auctions and spot markets.

Result 3: Tight holding limits reduce efficiency relative to loose holding limits. The top row of Table 3 reports the results of the (one-tailed) rank-sum test of the null hypothesis that the tight holding limit has no effect on efficiency. We can reject the hypothesis of no effect at the 3.8% level of significance, a result confirmed by our OLS estimates reported in Table 4. Tight holding limits reduce allocative efficiency in our experimental setup. Taken together, these results give weight to the argument that tight holding limits may impose considerable costs by interfering with the efficient allocation of allowances between sources and across time. Firms in the market face higher risk and receive less accurate signals concerning where resources should be invested in emission reduction activities. Since one of the justifications for the holding limits is to avoid market manipulation, it is especially noteworthy that tight holding

Fig. 8. Price dispersion each period by holding limit.

Please cite this article in press as: Shobe, W., et al., Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.04.007

G Model

ARTICLE IN PRESS

JEBO-3346; No. of Pages 19

W. Shobe et al. / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

15

Fig. 9. Average surplus lost each period.

limits, by lowering liquidity, tend to operate directly counter to one of the key defenses against market manipulation. We conclude that tight holding limits may make the allowance market more vulnerable to some of the commonly discussed strategies for market manipulation. What we do not yet know is whether any protections provided by limiting the share of allowances owned by one or a few large providers outweighs the costs and risks imposed by the holding limits themselves. In the least, it does seem clear that the holding limits should be no tighter than necessary. 4.2. The price containment reserve A price containment reserve is a stock of allowances that only becomes available for use in the market once some high price trigger-point is reached. RGGI, for example, has chosen a reserve that becomes part of the supply of allowances available at auction whenever the auction would otherwise close above a trigger price. The allowances available at that price point would be released to accept additional bids until the supply of allowances available at that price is exhausted or the auction closes, whichever comes first. California’s approach establishes a post-auction sale where potential buyers may bid for allowances at any of three price points. Because of the different release mechanism and the somewhat complicated incentive structure in the California reserve, it is not clear how participants in the California market will respond to the presence of the reserve or what would be its likely effect on price. We tested whether the different PCR designs produce significantly different permit market outcomes. While our results show that there are sharp differences in the way that the different reserves were used by the subjects in our experiments, it turns out that these differences do not translate into measurable differences in liquidity, price, price volatility or efficiency. This was true for our rank-sum tests on session-level data and for our subsequent panel regressions. For these tests, we used only observations from our Sessions 8, 10, 12, and 14 through 18. These sessions all had tight holding limits, but four used the post auction sale and four released permits into the auction when the price would otherwise close higher than the trigger prices. Both treatments had three price tiers of $23, $28, and $33. The reserve contained 36 permits or just over 7% of the total cap of 498 for the 12 period session. The rank-sum test results appear in the second row of Table 3 and clearly show the null hypothesis of no effect of the PCR design on market outcomes cannot be rejected. This is confirmed by the regression results reported in Table 4, where the estimated coefficient on the Post auction reserve sale variable is highly insignificant. The post auction reserve sale, while changing the temporal distribution of allowances released, does not have a measurable effect on any of our market outcomes. Result 4: The choice of post auction or in auction price containment reserve sale does not have a measurable effect on market outcomes. There are many possible explanations for the invariance of market outcomes to the price containment reserve treatment. One simple explanation is that the aggregate supply of allowances does not change as between the two and that it is the aggregate supply that matters rather than the way the allowances are given out. By this explanation, participants would be expected to use allowances the same way regardless of when they are made available. This explanation does not appear consistent with our earlier result that, during tight holding limit sessions, participants are clearly constrained in the extent Please cite this article in press as: Shobe, W., et al., Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.04.007

G Model JEBO-3346; No. of Pages 19

16

ARTICLE IN PRESS W. Shobe et al. / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

Fig. 10. Permit purchased in the post auction PCR sale by holding limit.

to which they can time-shift the available supply of allowances and that this has a substantial influence on the path of prices and production. Other possible explanations are that the reserve is just not large enough to make a detectable difference or that the prices are set so high that not enough of the reserve is realistically available to the market. Based on the information we have, we are not in a position to explain why the reserves as we modeled them in our experiments did not have a noticeable affect on market outcomes. We can, however, point to some differences in the way the reserves are used. These differences may point toward future explorations concerning the design of price containment reserve mechanisms. From a participant’s point of view, there may be advantages to having an opportunity to choose when to purchase allowances from the reserve as opposed to only purchasing them when some overall market trigger-point is passed, even if the individual would expect to use the allowances at the same time. Allowances that can be purchased by individual choice (even at a high price) provide a kind of insurance that is not provided by a general relaxation of the fixed market supply curve. An individual participant who expects increased scarcity in the future knows that auction and spot market prices can be expected to be high during those periods, and there is risk that bids on permits essential for compliance will be rejected should auction prices spike more than expected. With the post auction reserve sale, individuals may purchase allowances in advance at the fixed, if high, price available in one of the price tiers. The more likely it is that some period will have a price spike that will rise above the tier price, the more profitable advance purchase will seem. Based on this perspective, we would expect to see purchases from the lower price tiers well before the market price reaches the trigger price for that tier. We should observe more of these anticipatory purchases in tight holding limit treatments than in loose holding limit treatments, because the participants should expect greater shortages and higher spikes in price. From the point of view of the policy maker managing the program, the obvious risk in this anticipatory purchasing is that the reserve may be drawn down in advance of need to a point where it will not serve to reduce price spikes during periods of high demand. Fig. 10 shows the pattern of purchases from the post auction price containment reserve sale in Sessions 7 through 14 of our experiments. The top panel shows purchases under a loose holding limit and the bottom panel a tight holding limit. One would expect to see relatively large purchases during the high demand periods that we label as “low hydro.” These high demand periods, 4, 7, and 11, do show modest purchases, as expected in both tight and loose treatments. But just as many permits were purchased in the first three periods from the $23 tier of the reserve, when the auction closing price was very close to the reserve price of $10. This appears to be speculative buying in anticipation of substantial future price increases. These buyers are correctly anticipating that prices will eventually rise above the $23 trigger price. Anticipatory buying is larger on average in the loose holding limit sessions than in the tight ones. This seems quite reasonable. In the loose holding limit sessions, these allowances may be held in trading accounts rather than having to go into the non-tradable compliance accounts. It is also interesting to note the number of permits purchased from the reserve in the years after a low hydro period as participants in the tight holding limit sessions restock depleted banks of permits. There is a clear interaction between the tightness of the holding limit, and the likelihood that firms will buy allowances in the post auction sale. If this incentive is strong enough, the reserve could become depleted, as occurred in one of our tight Please cite this article in press as: Shobe, W., et al., Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.04.007

G Model JEBO-3346; No. of Pages 19

ARTICLE IN PRESS W. Shobe et al. / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

17

Fig. 11. PCR permits released post auction (top) and in auction (bottom).

holding limit sessions when the entire reserve was used up before the end of the session. Participants in tight holding limit sessions tend to draw more on the reserve in later periods because they are unable to keep the price from rising due to their limited ability to bank allowances. For emitters, the consequences of non-compliance go beyond the financial penalties of acquiring additional allowances. Firms and other institutions receiving penalties face substantial costs of bad publicity as well. It is reasonable to expect that this prospect would cause many market participants to be quite risk-averse toward a compliance penalty. Having the PCR available as a last protection against non-compliance may allow risk averse participants to bid closer to their actual values for allowances due to their smaller risk of inadvertent non-compliance. Fig. 11 shows the levels of allowances sold or released from the reserve by period, but depending on whether the sale was post auction (top panel) or in auction (lower panel). This is the division tested in our price containment reserve treatments. The different pattern shows clearly the important difference one would expect to observe in the release of allowances from the California and RGGI reserves. The reserve design in these programs roughly corresponds to the difference between our treatments. Allowances are only released from the reserve during price spikes with the in auction reserve but are released in anticipation of future scarcity in the post auction sale. On average 18.75 permits were released in post auction sale sessions and 12.75 permits released for the in auction sessions. For the in auction sessions, all allowances released from the reserve were released during the last two of the three low hydro years, while for the post auction sale, the use of allowances is spread across periods. It is not yet clear whether the invariance of our market outcome measures to the type of release mechanism is due to some specific features of our experimental setup or may reflect some more general principle. We can see that different designs may have decidedly different implications for the pattern of release of allowances from the reserve, but if this does not necessarily imply differences in market performance, then there may be other reasons for preferring one type of mechanism over the other. The specifics of the implementation of mechanisms for releasing price containment reserve allowances into the market would appear to be a fruitful area for further research. That this distinctively different pattern of allowances being released from the reserve has so little apparent effect on liquidity, efficiency, or price is not unlike “the curious incident of the dog in the night-time.”19 It did nothing in the night-time and that was the curious incident. So it is here, the lack of noticeable effect of the reserve release mechanism merits a little additional sleuthing. 4.3. The auction price rule The choice of auction pricing rule does not have any measurable influence on the market outcomes in our experiments. The Wilcoxon rank sum tests for the Lowest accepted bid variable (Table 3) do not support rejecting the null hypothesis of no effect. The regression estimates reported for the Lowest accepted bid variable (Table 4) also fails to find a significant relationship between the auction price rule and any of our measures of market outcomes. That the choice between different price rules does not affect the efficiency of permit allocations confirms a result from recent experiments by Cramton et al. (2012) where, in the context of an ascending clock auction with two bidders purchasing one item, they found no effect of the pricing rule on the efficiency of the allocation.

19

From the short story, “Silver Blaze” in Arthur Conan Doyle, Memoirs of Sherlock Holmes.

Please cite this article in press as: Shobe, W., et al., Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.04.007

G Model JEBO-3346; No. of Pages 19

18

ARTICLE IN PRESS W. Shobe et al. / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

Result 5: The liquidity, price variability, and efficiency outcomes do not depend on whether the lowest accepted bid or highest rejected bid auction pricing rule is used. Where there are several people bidding on numerous units of a good at auction, this may not seem very surprising. As the numbers of bidders and units for sale both rise, the likelihood that changes in a bidders marginal bids will have any effect on price goes down. Ties between the lowest accepted and highest rejected bids should become more common, as was the case with many of the auctions in these experimental sessions. 5. Conclusion Emission markets are created by legislation and regulation. Elements of emission market design can have important implications for how the markets perform (Burtraw et al., 2009; Cason, 2010; Holt et al., 2007). Economic experiments provide a particularly effective tool for identifying the effects of a given design element on outcomes of interest, effects that may be difficult to identify empirically. In this paper, we evaluate three design features in the cap and trade program that plays a central role in California’s regulatory regime for limiting GHG emissions. We examine allowance holding limits, the post auction compliance reserve sale, and the lowest accepted bid auction pricing rule. We find that, in our laboratory setting, the holding limits result in a large reduction in market liquidity, which must be considered a perverse effect, since the limits were put in place to help prevent market manipulation. The allowance holding limits have the additional unintended consequence of increasing price volatility and reducing market efficiency. Because market manipulation is known to be facilitated by periods of low liquidity (Pirrong, 2009), the AB 32 holding limit has the unintended consequence facilitating common forms of market manipulation. This points to the need to assess what levels of concentration of ownership may give larger market participants damaging price-setting power so that holding limits, if present, can be only as binding as is necessary. Future research should address whether there are less costly mechanisms for limiting market power of large emitters. Improved price collar designs may help. Our experiments also demonstrate how the three tier price containment sale, by providing a significant, albeit expensive, additional supply of allowances, can give market participants a buffer against the worst consequences of low liquidity periods. Buying from the reserve in anticipation of future scarcity adds a small element of “borrowing” from future supplies, since these reserve allowances have no vintage and may be used in any period once purchased. It is interesting that two very different designs for releasing the reserve into the market do not have measurably different effects on key measures of market performance: liquidity, price variability, and efficiency. The different patterns of reserve use that arise from the two mechanisms, justify increasing attention to the design of price collar mechanisms. Under different conditions, such as tighter caps or larger reserves, differences between them might emerge. Finally, we find that, in the context of our laboratory setting, the choice between the lowest accepted bid and highest rejected bid auction pricing rules does not have a significant effect on any of our outcome measures. References Aldy, J.E., Stavins, R., 2012. The promise and problems of pricing carbon: theory and experience. J. Environ. Dev. 21 (2), 152–180. Bernard, J.C., Mount, T., Schulze, W., 1998. Alternative auction institutions for electric power markets. Agric. Resour. Econ. Rev. 27 (2), 125–131. Bodsky, R., Donato, D., James, K., Porter, D., 2012. Experimental Evidence on the Properties of California’s Cap and Trade Price Containment Reserve. Economic Science Institute. Chapman University. Boyd, J., Burtraw, D., Krupnick, A., McConnell, V., Newell, R.G., Palmer, K., Sanchirico, J.N., Walls, M., 2006. Trading cases five examples of the use of markets in environmental and resource. In: Oates, W.E. (Ed.), The RFF Reader in Environmental and Resource Policy. Resources for the Future, Washington, DC, pp. 56–65. Burtraw, D., Goeree, J., Holt, C.A., Myers, E., Palmer, K., Shobe, W., 2009. Collusion in auctions for emission permits: an experimental analysis. J. Policy Anal. Manage. 28 (4), 672–691. California, 2006. California Global Warming Solutions Act of 2006 Cal. Health & Safety Code §§ 38500 et. seq. West Publishing. Cameron, A.C., Gelbach, J.B., Miller, D.L., 2008. Bootstrap-based improvements for inference with clustered errors. Rev. Econ. Stat. 90 (3), 414–427. Cason, T.N., 1995. An experimental investigation of the seller incentives in the EPA’s emission trading auction. Am. Econ. Rev. 85 (4), 905–922. Cason, T.N., 2010. What can laboratory experiments teach us about emissions permit market design? Agric. Resour. Econ. Rev. 39 (2), 151–161. Cramton, P., Filiz-Ozbay, E., Ozbay, E., Sujarittanonta, P., 2012. Discrete clock auctions: an experimental study. Exp. Econ. 15, 309–322. Dinan, T.M., 2010. Managing Allowance Prices in a Cap-and-Trade Program. Congressional Budget Office, Washington, DC. Ellerman, A.D., Marcantonini, C., Zaklan, A., 2014. The EU ETS: eight years and counting. Working Papers. Climate Policy Research Unit, Florence. Friesen, L., Gangadharan, L., 2013. Environmental markets: what do we learn from the lab? J. Econ. Surv. 27 (3), 515–535. Goeree, J.K., Palmer, K., Holt, C.A., Shobe, W., Burtraw, D., 2010. An experimental study of auctions versus grandfathering to assign pollution permits. J. Eur. Econ. Assoc. 8 (2–3), 514–525. Harris, J.H., 2010. Report on Holdings Limits to the Western Climate Initiative Markets Committee, Alfred Lerner College of Business and Economics. University of Delaware. Hart, O.D., 1977. On the profitability of speculation. Quart. J. Econ. 91 (4), 579–597. Hausker, K., 1992. The politics and economics of auction design in the market for sulfur dioxide pollution. J. Policy Anal. Manage. 11 (4), 553–572. Holt, C., Shobe, W., Burtraw, D., Palmer, K., Goeree, J.K., 2007. Auction Design for Selling CO2 Emission Allowances Under the Regional Greenhouse Gas Initiative. New York State Energy Research and Development Authority, Albany. Jarrow, R.A., 1992. Market manipulation, bubbles, corners, and short squeezes. Journal of Financial and Quantitative Analysis 27 (3), 311–336. Linklaters, 2011. California Agency Unanimously Adopts Cap-and-Trade Regulations: A Review of Key Program Design Elements and Outstanding Issues. Linklaters’ Environment and Climate Change Practice. Montero, J.-P., 2009. Market power in pollution permit markets. Energy J. 30 (2), 1–28. Newell, J., Pizer, W., Raimi, D., 2013. Carbon markets 15 years after kyoto: lessons learned, new challenges. J. Econ. Perspect. 27 (1), 123–146.

Please cite this article in press as: Shobe, W., et al., Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.04.007

G Model JEBO-3346; No. of Pages 19

ARTICLE IN PRESS W. Shobe et al. / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

19

Perkis, D.F., Cason, T.N., Tyner, W.E., 2014. An Experimental Investigation of Hard and Soft Price Ceilings in Emissions Permit Markets. Krannert School of Management, Purdue University (Unpublished manuscript). Pirrong, C., 2009. Market Oversight for Cap-and-Trade: Efficiently Regulating the Carbon Derivatives Market. Brookings Institution. Putnin¸sˇ , T.J., 2012. Market manipulation: a survey. J. Econ. Surv. 26 (5), 952–967. Roberts, M.J., Spence, M., 1976. Effluent charges and licenses under uncertainty. J. Publ. Econ. 5 (3–4), 193–208. Schmalensee, R., Stavins, R.N., 2013. The SO2 allowance trading system: the ironic history of a grand policy experiment. J. Econ. Perspect. 27 (1), 103–122. U.S. Energy Information Administration, 2013. State-Level Energy-Related Carbon Dioxide Emissions, 2000–2010. Van Horn, A., 2013. The PEAR/UVA Study: Testing AB 32’s Market Design – Allowance Holding Limits, the APCR, CPUC Purchase Limits and the CA Allowance Strategy Calculator. Van Horn Consulting, Orinda, CA. Weitzman, M.L., 1974. Prices vs. quantities. Rev. Econ. Stud. 41 (4), 477–491.

Please cite this article in press as: Shobe, W., et al., Elements of emission market design: An experimental analysis of California’s market for greenhouse gas allowances. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.04.007