The progressive inefficiency of replacing renewable obligation certificates with contracts-for-differences in the UK electricity market

Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Contents lists available at ScienceDirect

Energy Policy journal homepage: www.elsevier.com/locate/enpol

The progressive inefficiency of replacing renewable obligation certificates with contracts-for-differences in the UK electricity market Derek Bunn n, Tim Yusupov London Business School, Regent's Park, London NW1 4SA, United Kingdom

H I G H L I G H T S







The asset performance risk of wind investments deteriorates with greater wind penetration. Green certificates may offer lower investment risk than feed-in tariffs. Detailed fundamental modelling reveals subtle asset performance risks for wind. Recent UK policy changes may be ill-founded.

art ic l e i nf o

a b s t r a c t

Article history: Received 5 July 2014 Received in revised form 22 November 2014 Accepted 1 January 2015

This paper looks at the emerging risk/return profile for new renewable assets as a conventional wholesale electricity market progressively decarbonises. Using a detailed fundamental model of price formation risks, under increasing replacement of fossil fuel facilities with onshore and offshore wind, we show that the risk return profile becomes less attractive over time, and may therefore need sustained and possibly increasing policy support. Furthermore, we show that green certificate trading may become progressively more attractive as a supplementary support to wholesale prices, compared to fixed feed-intariffs. This is because the increasingly negative correlation between renewable output and wholesale prices reduces its revenue risk compared to fixed feed-in tariffs, if other factors remain constant, and thereby improves conventional financial performance risk metrics. In particular, this suggests that the recent energy policy change in Britain to move away from green certificates and into contracts-for-differences may have been ill-founded. & 2015 Elsevier Ltd. All rights reserved.

Keywords: Electricity Risk Investment Green certificates Feed-in-tariffs Wind

1. Introduction In the post-liberalised electricity markets, the political motivation to force through rapid technological change aimed at low carbon power generation and greater sustainability has inevitably led to awkward compromises and expedient interventions in market design. Creating a market price for the externality of carbon emissions was a theoretically attractive and practical solution, as the EU Emission Trading Scheme (EU-ETS) demonstrated since 2005, but not as complete a solution as anticipated. Even before the EU-ETS price decay, following an over-supply of allowances in the economic recession post 2008, separate initiatives to encourage the development of renewable technologies were instigated in the EU (and elsewhere) for reasons of sustainability and economic stimulus. Together with further subsidies for energy efficiency and additional carbon taxes, the multiplicity of policies n

Corresponding author. E-mail address: [email protected] (D. Bunn).

tended to crowd out the singular role of a carbon market and further contributed to its depressed price levels (Blyth et al., 2009). As for the renewables themselves, again, a market based solution through mandated quantity targets and green certificate trading had the appeal of allocative efficiency with the market participants deciding how best to meet quotas and at what price. But, compared to alternative methods such as fixed feed-in tariffs (FiTs), green certificate trading apparently created higher transaction costs (Mitchell et al., 2006). Thus, an extensive amount of research has looked empirically at the relative successes of green certificates, FiTs, and other incentives for renewable energy, and this accumulated research generally concluded, by 2012, that FiTs had been more effective in promoting renewable innovation (Butler and Neuhoff, 2008; Hass et al., 2011; Verbruggen and Lauber, 2012). The relative simplicity for new entrants of a fixed price was generally identified as the key factor, with Woodman and Mitchell (2011) suggesting that the absence of price risk may lead to a lower cost of capital for FiTs compared to green certificates. But, as so much of this evidence

http://dx.doi.org/10.1016/j.enpol.2015.01.002 0301-4215/& 2015 Elsevier Ltd. All rights reserved.

Please cite this article as: Bunn, D., Yusupov, T., The progressive inefficiency of replacing renewable obligation certificates with contracts-for-differences in the UK electricity market. Energy Policy (2015), http://dx.doi.org/10.1016/j.enpol.2015.01.002i

2

D. Bunn, T. Yusupov / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎

had been taken from retrospective observations on the early stages of renewable energy penetration in various countries, it is an open question whether indeed FiTs offer lower investment risks as low carbon penetration approaches the deeper decarbonisation targets that the policies envisage achieving by 2030 and beyond. This question requires a model-based analysis and in this paper, our results surprisingly suggest that in the case of wind, at least, green certificates would progressively offer investors a risk/return investment profile preferable to FiTs, after the early innovation stages and as decarbonisation deepens. Deep decarbonisation targets for the power sector are becoming widespread. The EU envisages the power sector to be decarbonised by 2050, with some countries, e.g. Britain, seeking to substantially achieve it by 2030, and creating legislation with that in mind (Climate Change Act, 2008). Decarbonisation pathways for transforming the generation mix (EC, 2011; Eurelectric, 2009; HMG, 2010; National Grid, 2011) provide a technologically feasible basis for the setting of policies and incentives, but in Britain, as well as elsewhere, raise the question (Ofgem, 2009; DECC, 2010) of whether the established liberalised energy market design is “fit for purpose” (i.e., whether it will produce efficient prices and investment signals for a low carbon transformation). Indeed, by 2012, the UK Government (DECC, 2012) was persuaded that the market needed substantial reforms and in particular that in order to foster the development of renewables, the existing green certificate system needed to be replaced by FiTs for specific technologies. This was a departure of economic ideology and one whose analytical basis rested upon suggestions that the absence of price risk in FiTs would reduce the costs of capital and thereby stimulate more investment. In the analysis developed here, we show that prospective investment risk requires a much more detailed and computationally intensive modelling approach. A crucial aspect is how price and output risk interact within the usual financial investment risk metrics that need to be satisfied for investors to commit funds. Britain is therefore a topical case study in which to model the prospective investment performance of the various technologies envisaged in the low carbon pathways, and in those contexts to test a counter-proposition that as renewable penetration increases, the relative investment risks may favour green certificates over FiTs. Britain introduced its green certificate mechanism, the Renewable Obligation Certificates (ROCs), in 2002, the market prices of which were determined by the demand and supply of renewable energy in the wholesale market on an annual basis. Demand was created by an obligation on retail suppliers to cover a specified fraction, increasing yearly, of their sales with certified green energy; supply was provided by metered renewable generation (with an administered buy-out supply to meet any shortfalls). In the UK, the ROCs were initially technology neutral, thereby tending to foster the less innovative solutions, but the Government soon rectified this through a more discriminatory procedure of awarding different amounts of ROCs to different technologies. Also, in order to ensure that ROC prices did not collapse, annual targets were systematically set to ensure that buy-outs at the administered price would be required. By 2012, ROCs had become successful in supporting the development of onshore and offshore wind, so much so that the UK had become the largest developer of offshore in the world, and that over 2012/13 a record 40% increase in wind capacity1 and a 56% increase in renewable generation had occurred,2 placing the UK fourth amongst G20 countries in 2013 for total renewable investment (PEW, 2014). So, at a time, in 2013,

when the pace of investment in wind was gathering momentum, and the regulatory risk of FiTs was becoming a concern elsewhere in Europe (Spain3 and other EU counties had retrospectively reduced feed-in tariffs, and recommendations in Germany were to move away from FiTs towards more market based approaches4), it was remarkably controversial to see the British Government suggesting that ROCs needed to convert into FiTs in order to achieve the required levels of investment. The proposition advanced in this paper is that a significant component of risk for wind investors is intermittency and its complex interaction with price risk leads to an investment risk metric that may favour ROCs over FiTs, as decarbonisation progresses. Lenders and ratings agencies use various risk metrics to evaluate financial investment plans (CPI, 2011; Moody's, 2009), and as a proxy for these in this analysis we refer to capital coverage risk (the probability that annual net earnings do not cover financing costs). Thus, how price risk interacts with output risk to provide an annual coverage metric is a crucial concern. As decarbonisation progresses, we show that there is an increasingly stronger negative correlation between market clearing prices and output. This is the well-known merit order effect, as in Sensfuß et al. (2008), Obersteiner and Saguan (2010), Gowrisankaran et al. (2011), Hirth (2012), which may, furthermore, be amplified in a market with strategic players as their market power is greater at times of scarcity (Twomey and Neuhoff, 2010). In many regions around the world with high levels of wind generation, low prices commonly occur during very windy periods (e.g. Australia, North America, Germany, Denmark, Spain) and even negative prices are sporadic events. With ROCs, therefore, exposure to the market price of electricity means that, if this negative correlation between price and output is sufficiently large, the total revenue distribution may be less risky (in terms of the conventional measure of variance) than that implied by price risk and output risk independently. Indeed, depending upon the negative correlation, and the relative contribution of price and output risk, ROCs can become an effective hedge in revenue risk compared to FiTs. Evidently this presumes that in the comparison, ceteris paribus, ROCs and FiTs are both set administratively to provide the same average level of remuneration per MWh produced. We discuss this further and the implications of our analysis for hybrid schemes such as Premium FiTs in the concluding section. In the next section we therefore create a detailed market simulation model to analyse the emergence of the above proposition in a realistic setting. Furthermore, we do this in a context that tests the various pathway assumptions, as proposed by the UK government for decarbonisation to 2030, against a conventional financial performance risk metric as well as the usual risk neutral expected rate of return. We describe the simulation set-up and the computational learning algorithm that allows us to models the emergence of prices above marginal cost. The results demonstrate that the various low carbon investment trajectories, if they are subjected to the plausible financial performance risk criteria that lenders and credit ratings agencies usually impose, may not be feasible without steadily increasing public support. Furthermore we show that for low levels of wind investment, the fixed feed-in tariffs are less risky, but after a moderate amount of wind in the technology mix, the green certificates (ROCs) become less risky than the fixed FiTs. We finally discuss the implications of this for the evolution of renewable support policies.

1 http://www.renewableuk.com/en/news/press-releases.cfm/record-breakingyear-of-growth-for-uk-wind-energy. 2 https://www.gov.uk/government/uploads/system/uploads/attachment_data/ file/244726/renewables.pdf.

3 http://www.the-european.eu/story-2536/spain-in-energy-policy-reversalback-to-coal-gas.html. 4 http://www.oxera.com/getmedia/97a39b7c-e751-4e5c-b53f-701ead6131af/ Energy-market-reform-in-Germany.pdf.aspx?ext¼ .pdf.

Please cite this article as: Bunn, D., Yusupov, T., The progressive inefficiency of replacing renewable obligation certificates with contracts-for-differences in the UK electricity market. Energy Policy (2015), http://dx.doi.org/10.1016/j.enpol.2015.01.002i

D. Bunn, T. Yusupov / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎

2. Method The motivating observation for this research can be expressed in the following proposition: Proposition 1. : As decarbonisation progresses, if the correlation between the market clearing price and output for a wind generator becomes increasingly negative there exists a critical level beyond which the volatility of revenue risk is lower with, than without, price risk. This is easy to show analytically in simple terms5, but requires very detailed modelling to assess what the critical level of correlation might be in practice. Evidently, beyond this critical level, it implies that output risk and price risk from the wholesale market would have a mutually compensating effect. In other words, high wind output at lower prices and low output at higher prices would lead to less revenue risk (measured by variance) than variable output at a fixed price. Hence the expectation that ROCs may have lower revenue variance than fixed FiTs, assuming that both are managed to provide the same average level of subsidy. Further, we expect that as more wind enters the technology mix, the correlation between prices and wind output will become more negative. However, we need a detailed simulation model to evaluate this more precisely. 2.1. Data The core of this analysis is a detailed price formation model of the wholesale electricity market in GB, specified for half-hourly resolution and explicitly including the individual statistical performances of all 133 generating assets of size greater than 20 MW connected to the National Grid6 at the beginning of 2012. This includes 28 wind farms, with their historic performance, geographical locations and local wind speed data. Price formation is simulated in the usual fundamental way according to the intersection of a supply function offered by the generators with demand. This approach was extensively iterated to derive a fully stochastic analysis of prices and outputs for all the main generating units for selected years up to 2030 based upon the National Grid central scenario for decarbonisation. Using historic probability distributions for wind, demand, fuel prices, availabilities, as well as their correlations, we derive prices and outputs and thereby annual revenue distributions for each generating asset. This leads to metrics of financial performance risk for each asset in terms of annual capital coverage probabilities. The analysis includes asset portfolio ownerships and thereby models the potential effects of market power in moving price levels to those more reflective of an imperfectly competitive market. This is done using iterative computational learning to derive sustainable prices above marginal cost. The model was initially calibrated to 2011/12 and substantial effort was given to precisely modelling location-specific wind farm generation and its portfolio implications. Then, the market structure is progressively changed according to the main technology mix pathway, 2015–2030, as 5 Let, E(p), V(p) be the mean and variance of price; E(o),V(o) be the mean and variance of output; and r is their correlation. Then V(po) is the revenue risk, being E(p2o2)  E2(po) ¼ V(p)2V(o)2 þ E(p)2V(o)2 þV(p)2 E(o)2 þr4V(p)2V(o)2 þ 2r2V(p)V(o) E(p)E(o) Thus, Revenue risk will be less than Price risk if V(p) 4V(p)2V(o)2 þE(p)2V(o)2 þV(p)2 E(o)2 þ r4V(p)2V(o)2 þ 2r2V(p)V(o)E(p)E(o) which admits real-valued solutions (Goodman, 1960). In particular, as r tends to  1, then this condition approaches V(p)/E(p) 4V(o)/E(o), i.e. if the coefficient of variation of price is greater than that of output, which is intuitive and quite plausible in practice. 6 Smaller grid connected units were grouped into an “others” category and embedded, distribution level generation was included on the demand side.

3

specified by the National Grid and DECC, and the annual financial performance risks of onshore and offshore wind assets are calculated. The UK carbon price floor trajectory is included, as well as central assumptions about ROCs and FiTs. We subject these pathway scenarios to tests of investability under central estimates, and compare the investment risks from ROCs and FiTs as decarbonisation deepens. 2.2. Simulation model The model is therefore comprised of two parts, a Base Simulator and a Gaming Simulator. The Base Simulator takes fundamental data and provides a stochastic baseline set of data. This data is then transferred to a Gaming Simulator which, in turn, allows the larger market players to individually optimize the prices and capacities offered from their portfolios and thereby collectively move market prices above marginal cost according to the results of the computational learning algorithm. These are described in more detail as follows. 2.2.1. Base simulator The Base Simulator runs a Monte Carlo stochastic simulation. The annual analysis is the aggregation of 12 monthly simulations, based upon simulating a typical day for each month in detail. For each day, probability distributions are specific to every half-hour “period”. For example, an input wind-speed distribution can apply to every day in January from 14:00 until 14:30, but no other period or month. Because of this assumption, the resolution of the simulation is set by the number of simulated “scenarios”, each of which represents the potential outcome of a day within that month. Each period is independent, and so is each month. Whist this serial independence in the simulations means that dynamic constraints such as ramp rates and start-ups are not explicitly specified, this is a common assumption in longer term modelling, and for example is used in the detailed medium term capacity assessment modelling by Ofgem (2013). A single simulation of a period within a scenario is referred to as an “instance”. For each instance, the wholesale market model is solved as the intersection of demand with the full supply stack. The base case simulations (2012) therefore included 1 year of 12 months of 50 scenarios with 48 half-hours which totals 28,800 instances. This was then repeated for 2015, 2020, 2025 and 2030 according to the various pathway assumptions. The model was programmed in Python whilst data was analysed with the use of Python, PostgreSQL and Microsoft Excel. The basic principles followed in this probabilistic simulation of price formation from a supply stack of generating units are common to several studies of longer term price formation and risk (e.g. Poyry, 2009; Munoz and Bunn, 2013), although the level of detail in this formulation is substantially higher than usual. The input for the base simulator includes period-specific demand profiles for business days and non-working days, wind speed distributions, generating unit availabilities, interconnector imports/exports and asset ownerships. It also includes monthspecific fuel price distributions and correlations, European Union emission allowance (EUA) price distributions and ROC price distributions. The market structure includes every generator located in Great Britain, calibrated to their respective historic capacity submission profiles, as well as the basic attributes of ownership, the technology type, the maximum generating capacity, and all their empirically-observed price-setting characteristics. In addition, for wind farms, their geographical positions were needed to account for output correlations. The input data and distributions for calibration were from 2006 to 2012. Intra-year fuel price distributions and correlations were fitted as log normal to data from the ARA Coal futures, Brent

Please cite this article as: Bunn, D., Yusupov, T., The progressive inefficiency of replacing renewable obligation certificates with contracts-for-differences in the UK electricity market. Energy Policy (2015), http://dx.doi.org/10.1016/j.enpol.2015.01.002i

D. Bunn, T. Yusupov / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎

4

⎧ ⎫ ⎪ (Bi + Fij P j + Eij P EUA − R ij P ROC ) (1 + Mi ) ⎪ ⎬ Ci = min ⎨ ⎪ ⎪ ⎩ (−PCFD + ΔPDAH )(1 + Mi ) ⎭ for f in{Gas,Coal,Oil} for i = 1, …, n

3. 4.

5. Fig. 1. Empirical distribution of availability declarations (MEL) for the Cottam power station in 2012.

Futures, EUA spot and NBP day-ahead markets. The ROC price distribution was fitted as a log normal to OFGEM data. The demand distributions are specified for business and non-business days. The generator capacity distributions were approximated with discrete distributions of available and dispatched output, as we observed from the data that most generators favoured specific output profiles. Data are available on the Maximum Export Limit (MEL) that each generating unit declares as its available capacity to the National Grid as well as it Final Physical Notification (FPN) which is what the generator expects to deliver in each trading period (this is generally what it has sold in the wholesale markets).These data were extracted from the Balancing Mechanism reporting archive.7 Fig. 1 shows a typical example of power station behaviour, with this particular coal fired station (Cottam) consisting of 4 generating units and being owned by a large portfolio player at the time and. Each generating unit's marginal cost was estimated from a combination of fuel price data and the bids and offers submitted by generators to the real-time balancing mechanism. Because a generator will seldom have an incentive to bid higher or offer lower than its marginal cost, fuel-cost multipliers were extracted as interpolations between the bid maxima and the offer minima. The wind speed distributions, from MIDAS,8 were estimated from hourly averages for each month to match a Weibull distribution. The wind farm power curves were obtained by calibrating each wind farm's local weather station's wind speed against the specific farm's power output on an hourly level.

2.2.2. The simulation meta-code Each instance within the simulation is created from the process 1. Random exogenous variables are simulated. These include the Demand, each fuel price f (inter-correlated), EUA price, ROC price. 2. The n generating units submitted prices are calculated as

7

http://www.elexon.co.uk/wp-content/uploads/2012/11/bmra_sd_v18.0.pdf. 8 MIDAS Land and Marine Surface Stations Data (taken from the British Atmospheric Data Centre) http://badc.nerc.ac.uk/view/badc.nerc.ac.uk__ATOM__da taent_ukmo-midas

6.

where C is the cost per MWh, P is price, B is the base cost per MWh, F is the fuel cost multiplier per MWh, E is the total emissions per MWh, R the number of ROCs per MWh, M the mark-up and ΔPDAH the day-ahead price error.9 Generator capacities are simulated from their availability characteristics. Wind farm outputs are simulated according to the process outlined in the next subsection. This also includes accounting for embedded wind generators on the demand side as well as (implicitly) wind-chill in the demand level. The merit order supply stack is constructed from generating units and interconnectors ordered in ascending order of submitted price (from least to most expensive). However, nuclear is always assumed to be at the bottom of the stack, although its marginal cost is higher than wind. This ensures that nuclear output is not curtailed.10 The market is cleared by having all active players take the price of the most expensive active generating unit needed to meet demand. This is recorded as the clearing price. If the demand is higher than the cumulative capacity including pumped storage reserve, a blackout is recorded.

2.2.3. Wind farm simulation process Each wind farm output is simulated based on an associated weather station. Each weather station was chosen based on its proximity to each wind farm and on the availability of adequate hourly wind data. Some wind farms use the same weather station, but this is corrected (see below). The power curve characteristics of each wind farm include a cut-in wind speed, a full-capacity wind speed and a cut-off wind speed. Below is the meta-code for simulating this: 1. Weather station data is simulated from historic distributions. 2. Weather station data is calibrated empirically to simulate wind speed at grid-connected wind farm locations. Grid-connected wind farm wind speeds are then correlated in accordance with their distance from one another. Fig. 2 shows all pairs of correlation coefficients between the 28 wind stations, assessed from hourly data from 2006 to 2012, as a function of their distances apart. 3. Grid-connected wind farm power outputs are inferred based on simulated wind speeds and power curves. 4. The cumulative embedded wind output is assumed to be related to the cumulative grid-connected wind farm output, and therefore a multiple of the latter is subtracted from demand for that instance. 5. Demand is correlated to the cumulative grid-connected wind farm outputs to account for wind-chill. This is a much more detailed approach than is often undertaken in similar models where manufacturer specified wind farm power curves are typically used together with [NASA's] average land wind speed grid to extrapolate each wind farm's output distributions. In 9 The variance of this is estimated from APX market price data and reflects the basis error in CfDs whereby the contract reference price may differ from the market price obtained by the company. 10 This may induce a price formation error in a stack price formation model depending upon how inflexible generators are treated within the market rules.

Please cite this article as: Bunn, D., Yusupov, T., The progressive inefficiency of replacing renewable obligation certificates with contracts-for-differences in the UK electricity market. Energy Policy (2015), http://dx.doi.org/10.1016/j.enpol.2015.01.002i

D. Bunn, T. Yusupov / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎

5

Fig. 2. Wind speed correlations as a function of distance.

our model, each wind farm has its empirically estimated power curve associated with a local weather station. This has the advantage of producing more realistic power output distributions by calibrating weather-station wind speeds to those actual wind speeds experienced by the turbines and thereby extracting empirical power curves. This empirical approach avoids estimation biases due to turbines being positioned at higher altitudes and in optimized geographical terrains to achieve higher-than-average wind speeds, and of manufacturers publishing power curves that do not, for various reasons, reflect actual wind-farm outputs.

2.2.4. Gaming simulator The Gaming Simulator is designed to mimic the strategic decisions of major players' to manage capacity availability as well as the generating unit mark-up decisions for each instance. As with many agent-based power market simulations (Sensfuß et al., 2007), this process is done using the Roth and Erev (1995) computational learning algorithm. This is an iterative multi-agent search process based upon each agent maximising a profit function through the reinforcement of successful strategies, the reversal of unsuccessful ones and persistent exploration of the

Please cite this article as: Bunn, D., Yusupov, T., The progressive inefficiency of replacing renewable obligation certificates with contracts-for-differences in the UK electricity market. Energy Policy (2015), http://dx.doi.org/10.1016/j.enpol.2015.01.002i

6

D. Bunn, T. Yusupov / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎

strategy space. Similar computational learning applications to electricity markets include, for example, applications that address vertical integration (Micola et al., 2008; Bunn et al., 2010), auction design (Nicolaisen et al., 2001) and market power (Browne et al., 2012). A probability distribution across capacity and mark-up choices is sequentially refined in this way for each agent with the “capacity gaming” stage preceding the “mark-up gaming.” This sequential process reflects the observation that capacity decisions are usually made as part of the asset optimisation process, at longer lead times than the determination of final price offers and bids closer to real-time. The gaming simulation learns iteratively to improve performance on all Base Simulator scenarios by adjusting capacity and mark-ups. Each generator's mark-up decisions are specific to each instance and the capacity decisions are specific to each scenario. This means that each generating unit's mark-ups are independent for every period and scenario while a generating unit is only allowed to submit a single capacity for an entire scenario (day). The justification for capacity decisions being scenario-specific is based on an extensive analysis of the MEL distributions for each power station (the example of one being shown in Fig. 1). We observed that each generating unit's intraday available capacity profiles are for the most part uniform, although they vary day-today. On the other hand, mark-up decisions are instance-specific because traders appear to adjust offer prices more adaptively during the day than capacity. The importance of modelling the capacity availability process is illustrated in Fig. 3, which shows time series for daily demand and availabilities in the GB Market over four years, 2005–2008. During these years, summer wholesale prices were maintained close to winter prices, and it is clear that the margin between available capacity and demand was being collectively well managed to achieve this result. Further examination revealed that it was mostly through the large portfolio companies profiling the availabilities of their coal assets that this steady reserve margin and price level was managed; it is an open question therefore whether the large wind farms that will replace them will behave in the same way, especially if they are mostly

joint ventures. The Gaming Simulator output is a function of the generators' behaviours, the maximum allowable loss-of-load probability, the maximum allowable generator mark-up, the minimum generator mark-up, and the mark-up and capacity profiles. While it is possible to allow all companies to be involved in the gaming process, only the sixteen largest players were activated as strategic, since these covered more than 95% of Great Britain's generating capacity in 2012. Other players are therefore price takers in the model and are assumed to submit stochastic capacities based on their historical profiles, as specified in the Base Simulator. Extreme capacity gaming that would result in blackouts is precluded by specifying a small maximum allowable loss-of-load probability. The exact value used is the highest pre-recession lossof-load probability in the sampled data for 2007. In other words, the model took the pragmatic perspective that resource adequacy will not drop below the pre 2008 recession level. If this value is exceeded all generators responsible for capacity withdrawal are heavily penalized in the model. In practice, these scenarios would not necessarily lead to blackouts because of the various reserves and contingencies that the system operator can utilise. As for limiting extreme gaming on prices, the maximum allowable price mark-up represents a form of behavioural barrier that limits players from setting unrealistically high mark-ups. All realistic gaming models for power markets without substantial demand elasticity need to be constrained with a price cap (Fabra et al., 2006), and we have used this as a calibration parameter to provide a good historical fit of the model to data in 2011 and 2012. The capacity profiles, which represent the specific capacities an agent can offer from each generating unit, are approximated by the set [0%, 50%, 100%] × A f where A f is the technology's availability factor. Mark-up profiles, on the other hand, are in the form of an exponentially-increasing interval set (e.g. [2%, 4%, 8%, … , M %] where M is the maximum markup). This form of mark-up profile allows for greater precision when competition is highest, whilst conserving the use of computing power. To the extent that the base load generators (wind, nuclear and, at times, coal) behave strategically, it will generally be through

Fig. 3. Time series for daily demand (lower) and available supply declarations (upper).

Please cite this article as: Bunn, D., Yusupov, T., The progressive inefficiency of replacing renewable obligation certificates with contracts-for-differences in the UK electricity market. Energy Policy (2015), http://dx.doi.org/10.1016/j.enpol.2015.01.002i

D. Bunn, T. Yusupov / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎

adjusting capacities (since they do not set the marginal price), whilst the marginal technologies (e.g. gas and oil) will tend to be strategic in mark-ups. This computational learning eventually settles on distinct asymptotic moving average values for each scenario capacity and each instance's clearing price for every strategic generating unit. In our analysis we typically needed 500 learning iterations to achieve acceptable capacity and price convergences for the 16 strategic companies, for each of the 28,800 instances created by the Base Simulator. The outputs of the Gaming Simulator are the mark-ups and capacity profiles, which, in combination with the Base Simulator output can then be used to infer each sample's supply function and the market-clearing price. The objective function is daily profit maximisation at the company level, and hence the 16 largest portfolio generating companies in the model learn, collectively through reinforcement (not collusive agreements), by means of iterated computational simulation, to adjust capacities and mark-ups of all their generating units to improve daily profits. The modelled price simulations were compared to actual APX power spot prices for the GB market as it was in 2011 and 2012. The simulated results reflected the empirical distributions sufficiently well for all months, after calibrating the strategic parameters. 2.3. Structural changes under decarbonisation We then used the model, calibrated as above to fit 2011/12, to represent the structural changes envisaged by the UK Government, as part of its obligations under the Climate Change Act (2008). The National Grid (2012) “Gone Green” pathway articulates this most precisely and our technology evolution, Table 1, is based upon that source. This provides the basis for extensive simulations of profit and load factor performance as well as market clearing prices. We assumed (initially) that decarbonisation retains the corporate market concentration level of 2011/12, with new investment entering pro-rata to plant retirements on a company basis. Furthermore, for new wind investments from 2015 onwards, we assume (initially) that they follow, pro rata, the same geographical distribution as the existing farms in 2011/12. The market values of ROCs are simulated based on their average price level (d45) and volatility in 2011/12. Their rate of Table 1 Central decarbonisation pathway (National Grid, 2012).

Generating Capacities (MW)Coal GAS (of which 4% is OCGT) Hydro Interconnector Nuclear Oil Pumped storage Offshore wind Onshore wind Embedded wind (as % of grid total) Solar PV Biomass Marine Total Embedded capacity Demand (2011: Avg,35GW; Peak 55,GW)% increase over 2011/ 2012 Fuels Brent ($/barrel) NBP (p/therm) ARA ($/tonne) Carbon (d) [UK floor price post 2012]

2011

2015

2020

2025

2030

27653.1 32852 908.03 3000 10555 3038 2930 659 821 35% 0 0 0 5700 0%

18180 30535 1247 4188 8980 0 2744 4925 4683 30% 898 2,355 0 6264  4.90%

17853 34773 1722 6588 9456 775 2744 16692 9167 16% 2500 2127 20 7175 0.65%

10042 40693 1752 7588 12121 606 2744 29842 10554 14% 4160 2417 329.5 8228 2.11%

10102 39185 1752 8588 13910 606 2744 36406 12041 13% 5801 2417 1081 9214 5.70%

111 56.4 121.5 11

110 72 90 20

128 82 104 30

148 92 121 50

171 102 140 70

7

allocation varies by technology and over time and for 2012, offshore wind received 2 ROCSs per MWh, whilst onshore received 0.9. The alternative FiT subsidies proposed by DECC (2013a, 2013b) are based upon administered strike prices for a Contracts for Differences (CfDs) support mechanism. This mechanism allows a renewable technology (and nuclear) to pre-contract or trade in the wholesale market, but subject to government-backed financial hedges (CfDs). The CfDs provide payments to (from) the generator if the reference market clearing price is below (above) the strike price. So there is clearly output (in terms of MWh generated) risk to the generator and also some price “basis risk” to the extent that the reference price formula (involving day ahead average prices) may not be exactly matched by the actual selling prices achieved by the wind generator. We assume in our central scenario that all new investment from 2015 onwards is with CfDs. For the CfD strike prices we have taken, from DECC (2013b), new investments at d155/MWh for offshore and d/100/MWh for onshore. For the initial simulations, we have not taken any degression projections of ROC and CfD, but as they are expected to reduce over time in line with decreasing capital costs, we undertake some sensitivity analyses on subsidy degression and capital costs later. The central capital costs estimated in DECC (2013a) are d1500/kW for onshore wind and d2500/kW for offshore. For assessing the investment risk, we look at the simulated probability distribution for annual capital coverage ratio. This is computed as the ratio of the annual net operational profit contribution of an asset divided by its annuitized capital cost. For onshore and offshore wind we assume financial lives of 20 and 30years, respectively and a cost of capital of 9% (pre tax).11 This coverage ratio is similar to the widely used debt coverage ratio (CPI, 2011), and would be a debt coverage ratio if the investments were 100% debt financed. We are aware that in GB, onshore wind has typically been about 80% debt finance and offshore rather less, but since equity investors generally target a higher rate of return than lenders, our capital coverage ratio can be taken as a rather conservative proxy for an overall ex ante financial performance risk criterion. If the average ratio is greater than 1, it is comparable to a positive NPV criterion, but its main value is in identifying risk tolerance and so we look at the 5% level of the simulated distribution as a value-at-risk criterion. Typically, credit rating agencies and lenders look at a wide range of financial coverage ratios, and we have taken a capital coverage ratio of at least of 1.2 at 95% probability as a representative proxy for maintaining investment grade debt. Whilst this value is clearly discretionary in an absolute sense, for the comparative insights required in this study, the proxy is sufficiently realistic. Because of the structural simulation of the asset performance and price formation, we implicitly take account of the interrelationship of market prices and load factors in this risk metric.

3. Results 3.1. Central pathway price evolution Firstly, the simulations demonstrate the changing nature of price risk as the market decarbonises, with both a declining average and a change in shape. In our base year, 2011, we have the average price of d49.5/MWh and in Fig. 4 we see the familiar 11 In DECC (2013a), levelised costs are computed at 10%, but this is admitted as being arbitrary. In the same report (p50) pre-tax hurdle rates for offshore under CfDs are between 9.6% and 11.3%, whilst onshore is estimated at 7.9%. Actual operating lives are estimated at greater than 20 years, although financial calculations may typically be over shorter periods (15yrs) and so, in that respect, actual capital coverage rates are likely to be lower.

Please cite this article as: Bunn, D., Yusupov, T., The progressive inefficiency of replacing renewable obligation certificates with contracts-for-differences in the UK electricity market. Energy Policy (2015), http://dx.doi.org/10.1016/j.enpol.2015.01.002i

8

D. Bunn, T. Yusupov / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Fig. 4. Wholesale market price distribution in 2011 (colours relate to daily trading periods). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

positive skewness with a few spikes, which has been the characteristic shape of wholesale power prices for many years until the advent of renewables (Weron, 2006). By 2015 this becomes more pronounced as some of the mid-merit plant retires. The average price of d61.1/MWh produced by the model is consistent with DECC (2012) projections. By 2020 (Fig. 5), whilst the mean price continues to increase (d67/MWh), we see the skewness change from positive to negative and emergence of negative price spikes (as had already become quite common in Germany, Denmark, the USA, Australia and elsewhere by 2012). By 2025, the negative prices start to give the shape a multimodal characteristic with prominent negative modes at the subsidy levels for onshore and offshore CfDs. The average price now starts to come down (d45/MWh). In these initial simulations, we have assumed that nuclear and

wind technologies would not withdraw capacity strategically. Nuclear is usually considered inflexible and to the extent that large wind farms may be owned as joint ventures, then this nonstrategic behaviour is plausible. But as part of large portfolio generators with market power, we would expect that wind, if not nuclear, would operate strategically. If this is the case, then we found that the average price could be maintained at d62.1/MWh. In our simulations for 2030, this effect is much more evident. Average price is now d14.9/MWh if wind and nuclear are not operated strategically, but if we presume that the wind farm operators will manage capacity strategically, then in Fig. 6 we again see average prices being held at d48.2/MWh. Observe in general that as renewable technologies compete for runtime through negative offers, it is the ones with the larger

Fig. 5. Wholesale market price distribution in 2020 (colours relate to daily trading periods). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Please cite this article as: Bunn, D., Yusupov, T., The progressive inefficiency of replacing renewable obligation certificates with contracts-for-differences in the UK electricity market. Energy Policy (2015), http://dx.doi.org/10.1016/j.enpol.2015.01.002i

D. Bunn, T. Yusupov / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎

9

Fig. 6. Wholesale market price distribution in 2030 with strategic wind capacity.

Fig. 7. Wholesale market price distribution in 2030 with strategic wind capacity and ROCs.

subsidies that can drive prices lower and still operate profitably. Hence the newer wind turbines, with the lower CfDs, will have the lower load factors. Perversely the more efficient technologies may be operating less frequently, and the higher the CfD strike prices the more the wholesale market becomes higher risk and lower return. With ROCs supporting the renewable technologies instead of CfDs, being set at the same average levels, and strategic gaming on wind, we have very similar average prices until about 2020, but thereafter under deeper decarbonisation, average market prices with ROCs becomes significantly higher as the negative tail is less pronounced. The average price in 2030 is d57.1/MWh, compared to d48.2/MWh, and the profile less risky (compare Figs. 6 and 7). Relating the above price distributions (CfDs without strategic wind) to the underlying wind speeds (weighted geographical average) in the simulations reveals a steadily increasing negative correlation. Fig. 8 reveals that in 2011, there was only a tiny,

insignificant negative correlation (  0.03) between market prices and wind. But, by 2030 (Fig. 9), this negative correlation had reached  0.7. The intermediate values were  0.16 for 2015,  0.41 for 2020;  0.64 for 2025. With ROCs they were very similar, and with strategic wind, slightly less. 3.2. Investability Table 2 provides the capital coverage ratio analysis for the central pathway, assuming the persistence of capital costs, CfDs and ROCs at the DECC (2013a, 2013b) levels. Looking at the P95 risk criterion,12 it is clear this gets steadily worse in all cases as decarbonisation deepens. Lenders and ratings agencies often have 12 This is a proxy for a conventional “value-at-risk” approach to investment whereby a value for the debt coverage ratio has a 95% chance of being exceeded.

Please cite this article as: Bunn, D., Yusupov, T., The progressive inefficiency of replacing renewable obligation certificates with contracts-for-differences in the UK electricity market. Energy Policy (2015), http://dx.doi.org/10.1016/j.enpol.2015.01.002i

10

D. Bunn, T. Yusupov / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Fig. 8. Wholesale market price distribution versus Wind Speed in 2011.

Fig. 9. Wholesale market price distribution versus Wind Speed in 2030.

a critical value for such ratios and as a proxy we have taken it as 1.2. We see for all 2030 simulations, it was below this critical value. In other words, without more subsidy, or interventions to reduce risk, lenders would assess the annual risk of not covering debt as increasing over time to possibly critical levels. It is perhaps surprising that CfDs become even more risky over time, compared to ROCs, but this is clearly due to the increased operational risk as wind becomes more marginal. In our simulations, we find that by 2030, onshore turbines are being curtailed 30%, and offshore 10%, beyond the load factors that the underlying wind conditions would provide. The reason that the model discriminates in this way between onshore and offshore is because of the higher subsidy that offshore wind is given, and therefore the higher opportunity cost is has for being curtailed. This leads to offshore undercutting onshore in conditions of oversupply when

wind is at the margin. Comparing onshore and offshore coverage ratios, in 2011, the average ratio with ROCs is higher for onshore, but as more wind get deployed, but by 2030, onshore performs worse than offshore in all the cases shown in Table 2. In general, ROCs are surprisingly less risky than CfDs, having higher P95 values and in some cases would support investments where CfDs would fail to meet the 1.2 criterion. It is also interesting to observe that when strategic capacity behaviour on wind is included, the offshore performs better (at the expense of onshore) and in 2025 will meet our reference investment criterion under ROCs, but not under CfDs. The simulation analysis has also provided results on all other technologies but these effects are really outside the scope of this paper. Briefly, however, all technologies, except open cycle gas turbines, suffered in terms of coverage ratio degradation as

Please cite this article as: Bunn, D., Yusupov, T., The progressive inefficiency of replacing renewable obligation certificates with contracts-for-differences in the UK electricity market. Energy Policy (2015), http://dx.doi.org/10.1016/j.enpol.2015.01.002i

D. Bunn, T. Yusupov / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎ Table 2 Capital cost coverage ratios for wind investment under the central decarbonisation pathway. (The values which would not meet an investment criterion 41.2 are in bold). Onshore CfD

(d1500/kw) ROC

Offshore CfD

(d2500/kw) ROC

Avg

Avg

Avg

P95

Avg

P95

capacity behaviour 1.49 1.37 1.63 1.51 2.13 1.6 1.42 1.98 1.13 0.81 1.3 0.87 0.57 0.91

1.98 1.8 0.96 0.64

1.44 2 1.97 1.35 0.98

1.38 1.87 1.78 0.99 0.66

1.8 1.72 1.1 0.81

1.44 1.86 1.85 1.5 1.23

1.38 1.75 1.74 1.24 0.89

P95

Without strategic wind 2011 2015 1.66 1.49 2020 1.52 1.41 2025 0.97 0.71 2030 0.68 0.48

P95

With strategic wind capacity behaviour: 2011 1.49 1.37 2015 1.35 1.24 1.36 1.25 2020 1.28 1.16 1.33 1.18 2025 0.92 0.66 1.09 0.87 2030 0.72 0.54 0.92 0.64

1.92 1.85 1.41 1.12

decarbonisation progressed, as a result of the price distributions creating higher risk. Whilst the ROC v CfD support mechanisms had little effect on the other technologies (since investment was not endogenised in this model), the degradation of financial performance as decarbonisation increases is clearly a factor that would have to be taken into account if capacity payment supplements, for example, are being considered to help new CCGT investments.

4. Discussion, conclusion and policy implications The impact of substantial amounts of renewable technologies creates a change in the nature of the wholesale electricity price formation process, and in particular they introduce a negative correlation between market clearing prices and renewable outputs. In itself, this has been well-known to hydro producers for many years, and more recently in the case of wind and solar. But with wind in particular changing the nature of fossil systems, this feature is causing asset performance risks to acquire new characteristics. Furthermore, this correlation increases strongly as more renewable generation replaces thermal facilities. Whilst it requires very detailed simulations to assess what this means for the impact of various policies on future investment risk, if these simulations are undertaken, the emergence of this negative correlation can be seen to have some risk mitigation effects. Thus, for example, if wind investments are exposed to wholesale prices, as well as being supported by supplementary green certificates, the revenue risk can, surprisingly, be lower than if it were insulated only from prices by a fixed feed-in tariff. In both cases, revenue depends upon output volume, but in the former case the negative correlation with prices acts as hedge. Evidently, this observation presumes that other factors remain constant in the comparison and that the feed-in-tariffs and green certificate values would evolve to give the same average prices per unit of output. Looking at the case of Britain, simulations suggest that in 2012, investment risk would be lower under a feed-in tariff (CfD) than through the pre-existing green certificate mechanism (ROCs), and so it is easy to understand why a policy change was proposed that year. But it appears to be too late, other things being equal, as by 2020, if wind investment proceeds as planned, the negative correlation between prices and wind output reaches  0.4, and this is sufficient to balance the two support mechanisms. Thereafter, as the correlation goes to 0.6 in 2025 and  0.7 in 2030, ROCs are

11

less risky, ceteris paribus, and may even be better able to meet riskadjusted investment criteria. Whether this is indeed how the relative merits of green certificates versus contracts-for-differences will be perceived by market participants is open to debate and there are many related factors that influence perceptions. Evidently, the analysis and simulations reported here focus upon one aspect, but they indicate that by itself, simply taking market price risk out of the investment case for new wind, will not necessarily make it more attractive. In practice, much depends upon relative attitudes to market and regulatory risks; administered feed-in-tariffs do carry regulatory risk, as evidenced by the distress caused to many companies in Europe when feed-in tariffs have been changed retrospectively,13 but so do green certificates (e.g. the policy reversals in Australia14 in 2014). The CfD mechanism in the UK promises index linking to general inflation, whilst ROC prices have been managed through quantities offered and annually administered buy-out prices. One can envisage various ways in which each could be mismanaged over time and for that reason, in our analysis, we assumed the interyear average remuneration per MWh would be set properly and equivalently for either ROCs or CfDs, and focussed on the intrayear stochastic drivers of annual financial performance. On balance, it appears that the upside potential of a ROC process may well compensate for the immunisation to the downside market price risks that CfDs provide. Furthermore, one of the benefits of a renewable obligation process is that there is an obligation on retailers to forward contract with renewable generators, which may induce a forward premium, whereas a contract-for-difference has no such investment pull from the retailers. Rather, a generator must offer into the market and seek to achieve at least the market reference price, as specified in the CfD formula. With regard to alternative forms of FiTs and market mechanisms, the insights from this study may generalise more restrictively, especially in markets where the price volatility is constrained by caps and negative price exclusions. But with hybrid support mechanisms, such as Premium FiTs to the energy market prices, our conclusions should extend, since the volatility of the ROCs had negligible impact in our main results. As for the model itself, a major concern is that the emergence of negative prices occurs in this model through the renewable generators receiving a subsidy and being willing to discount their offers up to the level of this subsidy in order to get dispatched. But, in several markets, the negative prices do not appear to emerge from this process. For example, in Germany, where the wind generators, supported by FiTs, do not need offer into the wholesale market to get dispatched, the negative wholesale prices are caused by inflexible conventional generators, or CHP facilities, seeking to avoid curtailments and expensive shut-down/start-up costs. Furthermore, these negative offers are often in excess of the subsidies to wind. Thus, it is likely that our model underestimates the negative tails of the price distribution quite substantially. On the other hand, our strategic parameters capped high price spikes in order to calibrate to 2011/2012 average price levels. This in practice, the price risks may have heavier tails, both upper and lower, than reflected in this modelling. But, as our simulations indicated that the case for CfDs becomes stronger than that of ROCs as the market risk increases, these biases in the tails should not reverse but would amplify our conclusions. 13 Retrospective changes to renewable tariff polices in Italy were cited in bond downgrades to below investment grade by Fitch Ratings (Bloomberg New Energy Finance Week in Review, Vol VI: Issue 256, 2014; http://www.bnf.com) 14 “AUSTRALIA DELIVERS POTENTIAL DEATH BLOW TO RENEWABLES”, Bloomberg New Energy Finance Week in Review, Vol VI: Issue 249, 2014; http:// www.bnf.com

Please cite this article as: Bunn, D., Yusupov, T., The progressive inefficiency of replacing renewable obligation certificates with contracts-for-differences in the UK electricity market. Energy Policy (2015), http://dx.doi.org/10.1016/j.enpol.2015.01.002i

12

D. Bunn, T. Yusupov / Energy Policy ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Thus, it does appear that the case in FiTs' favour is weaker than that of the more market based green certificate trading schemes, and that some conventional views on the risk mitigation advantages of FiTs do not hold up to closer analytical scrutiny.

References Blyth, W., Bunn, D.W., Kettunen, J., Wilson, T., 2009. Policy interactions, risk and price formation in carbon markets. Energy Policy 37, 5192–5207. Browne, O., Poletti, S., Young, D., 2012. Simulating market power in the New Zealand electricity market. N. Z. Econ. Papers 46 (1). Bunn, D.W., Matoccia, M., Ochoa, P., Kim, H., Ahn, N.-S., Yoon, Y.-B., 2010. Vertical integration and market power: a model-based analysis of restructuring in the Korean electricity market. Energy Policy 38, 3710–3716. Butler, L., Neuhoff, K., 2008. Comparison of feed-in tariff, quota and auction mechanisms to support wind power development. Renew. Energy 33, 1854–1867. Climate Change Act. 2008. 〈http://www.legislation.gov.uk/ukpga/2008/27/pdfs/ ukpga_20080027_en.pdf〉. CPI. 2011. The Impacts of Policy on the Financing of Renewable Projects: A Case Study Analysis. The Climate Policy Initiative. San Francisco. 〈http://climatepoli cyinitiative.org/publication/the-impacts-of-policy-on-the-financing-of-renew able-projects-a-case-study-analysis/〉. DECC. 2010. Electricity Market Reform Consultation Document 〈https://www.gov. uk/government/uploads/system/uploads/attachment_data/file/42636/1041electricity-market-reform-condoc.pdf〉. DECC. 2012. Electricity Market Reform: policy overview. 〈https://www.gov.uk/gov ernment/uploads/system/uploads/attachment_data/file/65634/7090-elec tricity-market-reform-policy-overview-.pdf〉. DECC. 2013a. Electricity Generation Costs 〈https://www.gov.uk/government/up loads/system/uploads/attachment_data/file/223940/DECC_Electricity_Genera tion_Costs_for_publication_-_24_07_13.pdf〉. DECC. 2013b. Levy Control Framework and Draft CfD Strike Prices. 〈https://www. gov.uk/government/uploads/system/uploads/attachment_data/file/209361/ Levy_Control_Framework_and_Draft_CfD_Strike_Prices.pdf〉. European Commission. 2011. Energy Roadmap 2050. COM (2011) 885/2. Eurelectric. 2009. Power Choices Report. 〈http://www.eurelectric.org〉. Fabra, N., von der Fehr, Nils-Henrik, Harbord, David, 2006. Designing electricity auctions. Rand. J. Econ. 37 (1), 23–46. Goodman, L.A., 1960. On the exact variance of products. J. Am. Stat. Assoc. 55 (292), 708–713. Gowrisankaran et al. 2011. Intermittency and the Value of Renewable Energy, NBER Paper 17086, 〈http://www.nber.org/papers/w17086.pdf〉. Hass, R., Resch, G., Panzer, A., Busch, S., Ragwitz, M., Held, A., 2011. Efficiency and effectiveness of promotion systems for electricity generation from renewable energy sources-lessons from EU countries. Energy 36,, 2186–2193. Hirth, l. 2012. The Market Value of Variable Renewables FEEM Report 15.2012 〈http://www.fee.it〉.

HM Government. 2010. 2050 Pathways Analysis. 〈https://www.gov.uk/government/ uploads/system/uploads/attachment_data/file/42562/216-2050-pathways-ana lysis-report.pdf〉. Micola, Augusto, Ruperez, Albert, Banal, Estanol, Derek, W. Bunn, 2008. Incentives and coordination in vertically related energy markets. J. Econ. Behav. Organ. 67, 381–393. Mitchell, C., Bauknecht, D., Connor, P.M., 2006. Effectiveness through risk reduction: a comparison of the renewable obligation in England and Wales and the feed-in system in Germany. Energy Policy 34, 297–305. Munoz, J.I., Bunn, D.W., 2013. Investment risk and return under renewable decarbonization of a power market. Climate Policy 13, 87–105. Moody's. 2009. Unregulated Utilities and Power Companies: Research and Ratings. 〈http://www.moodys.com/researchandratings〉. National Grid. 2011. Gone Green 2011 〈http://www.nationalgrid.com/NR/rdonlyres/ F6FA7970-5FEA-4918-8EE2-2A8E6B9626FF/50214/10312_1_NG_Future energyscenarios_factsheet_V2_st3.pdf〉. National Grid. 2012. UK Future Energy Scenarios 〈http://www.nationalgrid.com/NR/ rdonlyres/332FFA28-6900-4214-92BB-D3AD4FA5DC01/56611/UKFuture EnergyScenarios2012.pdf〉. Nicolaisen, James, Petrov, Valentin, Leigh, Tesfatsion, 2001. Market power and efficiency in a computational electricity market with discriminatory doubleauction pricing. IEEE Transactions on Evolutionary Computation 5 (5), 504–523. Obersteiner, C., Saguan, M., 2010. Parameters influencing the market value of wind power–a model-based analysis of the Central European Power market. Eur. Trans. Electr. Power 21 (6), 1856–1868. Ofgem. 2009. Project Discovery: Energy Market Scenarios. 〈https://www.ofgem.gov. uk/ofgem-publications/40361/discoveryscenarioscondocfinal.pdf〉. Ofgem. 2013. Electricity Capacity Assessment Report 2013. Ofgem, 2009. 〈https:// www.ofgem.gov.uk/ofgem-publications/〉. PEW Charitable Trusts. 2014. Who's winning the clean energy race?. Poyry, 2009. Impact of Intermittency: How Wind Variability Could Change the Shape of the British and the Irish Electricity Markets. Poyry Energy, Oxford. Roth, A.E., Erev, I., 1995. Learning in extensive form games: experimental data and simple dynamic models in the intermediate term. Games Econ Behavior 8 (1), 164–212. Sensfuß, F., Genoese, M., Ragwitz, M., Most, D., 2007. Agent-based simulation of electricity markets, a literature review. Energy Stud. Rev. 2007 15 (2.2). Sensfuß, F., Genoese, M., Ragwitz, M., Most, D., 2008. The merit-order effect: a detailed analysis of the price effect of renewable electricity generation on spot market prices in Germany. Energy Policy 36, 3076–3084. Twomey, P., Neuhoff, K., 2010. Wind power and market power in competitive markets. Energy Policy 38, 3198–3210. Verbruggen, A., Lauber, V., 2012. Assessing the performance of renewable electrcity support instruments. Energy Policy 45, 635–644. Weron, R., 2006. Modeling and Forecasting Loads and Prices in Deregulated Electrcity Markets. Wiley. Woodman, B., Mitchell, C., 2011. Learning for Experience? The development of the renewables obligation in england and wales 2002–2010. Energy Policy 39, 3914–3921.

Please cite this article as: Bunn, D., Yusupov, T., The progressive inefficiency of replacing renewable obligation certificates with contracts-for-differences in the UK electricity market. Energy Policy (2015), http://dx.doi.org/10.1016/j.enpol.2015.01.002i