Renewable Energy Support Policies and Wholesale ...

0 downloads 0 Views 644KB Size Report
a Georgia Institute of Technology; School of Economics, 221 Bobby Dodd Way, Atlanta, GA 30332; [email protected] (corresponding author).
Renewable Energy Support Policies and Wholesale Electricity Price Risk: A Stochastic Merit-Order Effect?† ERIK PAUL JOHNSONa AND MATTHEW E. OLIVERb a

Georgia Institute of Technology; School of Economics, 221 Bobby Dodd Way, Atlanta, GA 30332; [email protected] (corresponding author) b Georgia Institute of Technology; School of Economics, 221 Bobby Dodd Way, Atlanta, GA 30332; [email protected]

September 2016 Abstract: In deregulated electricity markets, economic support for renewable energy (RES-E) often shields investors from revenue risk arising from stochastic wholesale electricity prices. Two dominant RES-E support policies worldwide are feed-in tariffs (FIT) and renewable portfolio standards (RPS). We explore an under-researched benefit of these policies to retail electricity providers (and, ultimately, to consumers)—that stimulating investment in RES-E generation itself reduces wholesale price risk. We demonstrate that, in theory, greater RES-E generation should reduce the short-run variance in the wholesale electricity price. We refer to this as a stochastic merit-order effect. We find empirical support using a panel of policy, price, and generation data for 19 countries over the period 2000-2011. Both FIT and RPS are associated with reduced short-run variance in wholesale electricity prices. Moreover, we find evidence that FIT countries experience reduced electricity price variance over the long run, whereas with RPS results are mixed. JEL Codes: Q4, Q42, Q48, Q5 Keywords: renewable energy policy, electricity price risk, feed-in tariff, renewable portfolio standard



This research was presented at the 17th Annual CU Environmental and Resource Economics Workshop, the Association for Public Policy Analysis & Management (APPAM) 2015 Fall Research Conference, the 85th Annual Meetings of the Southern Economic Association, the 2016 Midwestern Economics Association Annual Meeting, and at seminars at Georgia Tech and Appalachian State University. We thank participants of those events for a number of helpful comments and observations. Additionally, we thank Wesley Burnett, Christoph Graf, Ian Lange, Mark Lively, Juan Moreno-Cruz, Mar Reguant, and Richard Schmalensee for many useful comments and suggestions. We are indebted to Nick Johnstone for providing the international RPS and FIT data. Chris Blackburn provided essential graduate assistant work related to data cleaning, and Mishal Ahmed performed minor graduate assistant work related to data collection. Finally, we thank the Georgia Tech School of Economics for financial support.



1

1. Introduction The transition from a global energy economy based on fossil fuels to one based on carbon-free renewable resources is among the most pressing and challenging issues of our time. In the electricity sector, most renewable technologies are not yet competitive with conventional fuels due to higher generation costs per kilowatt-hour (kWh). This disadvantage inhibits incentives for investment in renewable generation capacity. Many governments around the world have implemented economic support policies to stimulate investment in renewable generation, with the ultimate goal of reducing carbon emissions in response to increased public concern over the potential risks of anthropogenic climate change. Two dominant renewable support policies have emerged. Feed-in tariffs (FIT) guarantee all eligible renewable-energy-source electricity (RES-E) producers receive a fixed price (or fixed premium) per kWh generated, and obligate the nearest utility provider to purchase and distribute all available RES-E (Cory et al. 2009; Mendonça et al. 2010). By contrast, under a renewable portfolio standard (RPS), retail electricity providers are required to procure a specific minimum proportion of supply from renewable sources. These support policies have helped usher in a dramatic increase in (non-hydroelectric) RESE generation over the past 25 years, led primarily by wind and solar, along with a significant increase in geothermal energy and more modest increases in generation from other sources. According to our data, as of 1990 only Portugal had enacted a FIT for any RES-E technology. By 2011, the number of OECD countries with wind and solar FIT polices had increased to 24, and the number of countries with a FIT for geothermal had risen to 16; many also have FITs for other, less-developed RES-E technologies. The first RPS emerged in the U.S. a few years later in 1998. A total of eight OECD countries had enacted an RPS requirement by 2010.



2

Table 1. RES-E generation: 1990 versus 2013, with projection to 2040. OECD total (GWh). Sources: IEA (2014a,b); EIA (2016). Source Wind Solar PV/ thermal Geothermal Total Pct. of total generation

1990 3,844 681 23,190 27,715 0.36

2000 28,534 1,244 25,752 56,069 0.56

2013 435,854 111,136 33,973 581,922 5.35

2040 (proj.) 1,592,000 638,000 134,000 2,364,000 16.89

As shown in Table 1, the increase in RES-E generation over this period has been commensurate with these countries’ policy efforts, which serves as an indicator of the policies’ effectiveness in stimulating investment in RES-E generation. According to the International Energy Agency (IEA), in 1990 only 0.36 percent of OECD electricity generation came from wind, solar, and geothermal sources, but by 2013 the share had increased to 5.35 percent. This trend is expected to continue over the coming 25 years as support policies are renewed in some countries and introduced in others. The U.S. Energy Information Administration (EIA) projects that by 2040 16.89 percent of OECD electricity generation will come from these three RES-E technologies alone. A robust literature spanning disciplines such as economics, policy, and electrical systems engineering has sought to compare and contrast RPS and FIT on multiple dimensions. We contribute to this literature by exploring an under-researched aspect of the RPS-FIT discussion. Specifically, our main goal is to empirically examine the short-run variation in wholesale electricity prices—a fundamental source of risk in electricity markets—that emerges under RPS and FIT schemes. It is well known that deregulated electricity markets are prone to significant variability in wholesale prices, resulting from a number of factors including variation in fuel



3

prices, availability of generation capacity, unexpected outages, demand elasticity and exogenous demand variations, the lack of large-scale storage capability, and transmission constraints (Benini et al. 2002). Stochastic price fluctuations, compounded by the ‘intermittency’ problem associated with key renewable technologies like wind and solar, imply risk and uncertainty are unavoidable aspects of the renewable generation problem.1 Although the lack of cost competitiveness of RES-E is a motivating factor for policy makers when considering RES-E support schemes, other proponents have argued that such policies have the additional benefit of insulating investors from revenue risks associated with electricity price variability. But investors are not the only market participants to whom a reduction in price risk might be beneficial. Retail electricity providers mitigate wholesale price risk through hedging, futures markets, and other costly risk management strategies, but are unable to perfectly shield themselves from price risk. Thus, the costs of such risk faced by electrical utilities must ultimately be borne in the form of higher risk premiums passed on to electricity consumers. Intuitively, in the short run any policy that stimulates RES-E generation should be associated with a reduction in the variability of wholesale electricity prices paid by utilities. We refer to this reduction in price variability as the stochastic merit-order effect. As we explain below, a policyinduced increase in RES-E generation shifts the electricity supply curve outward, implying the stochastically fluctuating demand curve intersects it at a flatter section, suppressing the resulting price variability (Johnson and Oliver 2016). Our theory follows from a stochastic extension of the well-known merit-order effect of RES-E generation capacity. The merit-order effect states that as long as the electricity supply curve is upward sloping (or step-wise increasing), the 1

Some have argued that greater wind penetration might lead to increased price volatility. When conventional generators have market power, electricity prices may be further depressed during times of excessive wind output as conventional generators buy back excess supply, whereas they are able to increase prices further during times when output is low (Twomey and Neuhoff 2010).



4

reduction in conventional electricity demand resulting from increased RES-E capacity will reduce the spot market price of electricity. The merit-order effect has been observed as an indirect result of RES-E support policies in Germany (Sensfuß et al. 2008; Tveten et al. 2013; Cludius et al. 2014), Spain (Sáenz de Meira et al. 2008; Azofra et al. 2014), and Italy (Clò et al. 2015).2 We find support for our theory by analyzing an unbalanced panel of policy, price, and generation capacity data for 19 countries over the period 2000-2011. Specifically, we regress the (log) quarterly variance of daily electricity prices on FIT and RPS policy variables, controlling for other relevant covariates. We find both FIT and RPS are associated with a reduction in the variance in prices, which we conclude is driven primarily by the stochastic merit-order effect. We also examine how the presence of RES-E support policies may affect the long-run variance in electricity prices. Using cross-country variation in trends in the variance of electricity prices, we find limited evidence of a long-run stochastic merit-order effect for countries with FIT. For RPS we are generally unable to reject the null hypothesis of no long-run effect on perunit electricity price variance. Two closely related papers to our own find somewhat conflicting results. First, Woo et al. (2011) empirically estimate the impact of wind generation capacity on intraday prices in Texas. Their primary result is the merit-order effect—greater wind generation capacity is found to reduce intraday spot prices across ERCOT’s four zonal markets. Woo et al. then use their parameter estimates to compute the effect of a ten-percent increase in wind generation capacity on price variance. They predict increases in intraday price variance of roughly five percent in the ERCOT-West zone and less than one percent in the other three zones, arguing that if expanded

2

Nelson et al. (2013) examined the existence of a merit-order effect related to solar FITs in Queensland, Australia, finding the effect to be transient and not welfare enhancing.



5

wind-generation capacity has a “large impact” on electricity price variance, it may require increased use of electricity price risk-management techniques. However, it is hardly clear whether increases in price variance of less than one percent should be considered “large” (or even statistically significant), or whether the ERCOT-West zone, with a more substantial predicted price variance increase of five percent, is sufficiently representative of a normal electricity market for wider inference to be drawn. Wozabal et al. (2014) develop a similar theoretical model to ours, examining the effect of intermittent energy sources on electricity price variance in the German power market. Wozabal et al. provide a more direct test of the effects of RES-E on price variance. Specifically, they use intraday price variance as their dependent variable (unlike Woo et al., who use price as the dependent variable and then use their coefficient estimates to back out predicted changes in price variance). We choose to follow Wozabal et al. by using the (log) quarterly variance in wholesale electricity prices as the dependent variable in our regressions. In contrast to Woo et al. (2011), Wozabal et al. find that increased production of intermittent generation generally reduces wholesale price variance, although the opposite occurs for “very low and very high” levels of intermittent generation relative to total demand (the latter being a likely explanation of the Woo et al. (2011) prediction for the ERCOT-West zone in Texas). Ultimately the effects depend on the distribution of intermittent generation and the slope of the supply function, which complements our own theoretical predictions and empirical findings. Given the relatively sparse literature on the relationship between RES-E investment and electricity price variance—and the apparent lack of consistency in the results of prior studies of single markets at the intraday level—we offer an alternative framework by testing for effects at the daily level across markets. We further distinguish our work from these related papers by



6

testing for differential effects by policy choice. Our motivation here is that because FIT is a price-based instrument, whereas RPS is a quantity-based instrument, it is possible—even likely—that these two alternative policies affect electricity price variance in different ways. As we explain later when presenting our results, this intuition is borne out by our coefficient estimates in an interesting way. The remainder of the paper proceeds as follows. Section 2 provides a brief overview of RPS and FIT policies. In section 3, we describe the simple theoretical intuition for the stochastic merit-order effect as it relates to RES-E support policies. Section 4 describes our data, empirical design, and estimation results, and provides further discussion of the implications for policy and industry. We also present an alternative specification as a robustness check. Section 5 provides a limited test of the existence of a long-run stochastic merit-order effect. Section 6 concludes. An Online Appendix is available that (i) presents our short- and long-run analyses using the variance in electricity expenditures per unit as the dependent variable (that is, a weighted-average price, accounting for policy-specific payments to RES-E), and (ii) provides more detailed information on our data sources.

2. Policy Overview: FIT versus RPS Many studies have explored the advantages and disadvantages of FIT and RPS using various objective criteria.3 While an exhaustive review is not warranted here, we outline below what we believe to be fundamental features of the two policies.

3

Many have compared and contrasted FIT and RPS with other RES-E support schemes, including investment tax credits, production subsidies, clean energy standards, net metering, carbon emissions taxes, carbon cap-and-trade, bidding auctions for long-term purchase contracts, and others (e.g., Madlener and Stagl, 2005; Palmer and Burtraw, 2005; Huber et al. 2007; Finon and Perez 2007; Mulder 2008; Timilsina et al. 2012; Fell and Linn 2013; Johnson 2014). Some researchers have also begun to study the implications of overlapping RES-E support policies (e.g. Cory et al. 2009; Fischer and Preonas 2010; Böhringer and Rosendahl 2010).



7

FIT supports investment in RES-E in two ways.4 First, it guarantees all eligible producers receive per kilowatt-hour (KWh) a fixed price or the spot price plus a fixed premium (Cory et al. 2009). Second, the nearest utility provider is obligated to purchase and distribute all RES-E that ‘feeds-in’ to the grid, regardless of electricity demand (Mendonça et al. 2010). Tariff levels are typically based on a generator’s levelized cost-of-service (Couture and Cory 2009). Total generation costs per kWh vary across technologies and sites, and include the costs of capital investment, regulatory compliance and licensing, operation and maintenance, fuel costs (for biomass and biogas), inflation and interest, and a rate of return on investment (Klein et al. 2010). The most commonly used remuneration period is 15-20 years, where 20 years is considered the average life of a typical renewable energy plant (Mendonça et al. 2010). Importantly, most FITs provide for (i) ‘tariff digression’; and (ii) tariff review and revision. Tariff digression is defined as a level of remuneration that depends on a plant’s vintage—newer plants receive lower guaranteed payments, increasing the incentive to install new capacity sooner rather than later and stimulating technological improvement. The possibility of review and revision reflects an acknowledgment of underlying technological and market developments that may unexpectedly affect capacity costs due to input price shocks (Klein et al. 2010). RPS is a quantity-based instrument in which the regulator requires that a specified minimum proportion of electricity come from RES-E (typically per year). Electric utilities can meet RPS requirements by purchasing RES-E from independent generators, or through the installation and operation of their own facilities (Wiser et al. 2005). RPS policy typically includes a complementary market for tradable renewable energy certificates (RECs).5 For every megawatthour (MWh) of RES-E generated, a REC is created. The utility pays the renewable generator for

4

For a complete overview, see Mendonça et al. (2010). A more concise, but informative review of alternative FIT design options is available in Couture and Gagnon (2010). 5 Also commonly referred to as ‘tradable green certificates’.



8

both the electricity supplied and the REC, providing renewable generators with a supplemental income stream. Each year, RECs are surrendered to the jurisdictional regulator to demonstrate compliance with the RPS. Alternatively, RECs can usually be ‘banked’ for future use (Johnson 2014). Utilities with RECs in excess of the RPS requirement can sell them in the market for RECs, while others might purchase RECs as a substitute for purchasing electricity directly from renewable generators (Wiser and Barbose 2008).6

3. The Stochastic Merit-Order Effect To provide intermediate-level microeconomic intuition for why RES-E support policies might be expected to reduce short-run electricity price risk, we recap the theory proposed in Johnson and Oliver (2016). Consider the simple diagrammatic model in Figure 1. Let 𝐷(𝑃) be the inverse electricity demand curve, where 𝑃 is the wholesale price of electricity. Assume 𝐷(𝑃) has some stochastic component (related to weather, for example) that generates positive and negative demand shocks in the short-run. For simplicity, define 𝐷 𝑃 as the upper limit for a positive short-run demand shock and 𝐷(𝑃) as the lower limit for a negative demand shock. Let 𝑄 denote the quantity demanded/supplied. Assume the short-run supply curve for electricity, 𝑆(𝑃), is relatively flat for low supply quantities, but rises sharply as 𝑄 approaches maximum generation capacity, 𝑄. This is consistent with the conventional wisdom concerning short-run electricity supply curves. Panel (a) depicts the baseline scenario with no RES-E support policy, which we assume results in zero RES-E generation. Given the upper and lower bounds of the stochastic demand curve, the equilibrium wholesale price fluctuates between 𝑃∗ and 𝑃∗ . In panel (b), the RES-E

6

See Amundsen and Mortensen (2001) for formal analytical treatment of the quota system (RPS) with tradable green certificates (RECs).



9

(a) Without RES-E support

(b) With RES-E support

Figure 1. Simple model of an electricity market. Source: Johnson and Oliver (2016)

support policy induces RES-E generation amount 𝑄( , shifting the electricity supply curve to the right, from 𝑆(𝑃) to 𝑆(𝑃), and effectively increasing maximum generating capacity from 𝑄 to 𝑄 = 𝑄( + 𝑄. With RES-E generation, in the absence of a demand shock, the equilibrium wholesale price of electricity falls from 𝑃∗ to 𝑃∗∗ —the merit-order effect. Additionally, the range of variation in the wholesale price is lower, fluctuating between 𝑃∗∗ and 𝑃∗∗ . This reduction in variation is the stochastic merit-order effect. Thus, we expect that any policy leading to a shortrun increase in RES-E generation should result in lower variability in the wholesale price, driven by movement downward along the electricity supply curve.7 Note that the same argument holds even if 𝑄( is stochastic because of intermittency.

7

Note also that the stochastic merit-order effect implies negative demand shocks should lead to a greater truncation of negative price fluctuations than positive ones. This effect is referred to as the ‘inverse leverage’ effect, for which Knittel and Roberts (2005) provide empirical evidence.



10

4. Empirical Analysis 4.1. Description of the Data We empirically analyze the relationship between RES-E support policies and electricity price variance using wholesale electricity market data and national RES-E policy support levels. The final sample used in our estimation consists of an unbalanced panel of 19 countries over the period 2000-2011. Our data were compiled from multiple sources, rendering the data-cleaning process a formidable one. The reward is that our dataset is entirely unique to this paper. FIT and RPS data are taken from Johnstone et al. (2010), and contain FIT payments (by resource) and RPS requirements for each country at an annual interval. U.S. REC prices were purchased from Marex Spectron;8 all others are publicly available on the web. Figure 2 displays average FIT payments by resource by year across countries, RPS requirements by country by year, and REC prices by country by year. Daily and/or hourly wholesale electricity prices were in some cases publicly available (e.g., U.S., Canada, Australia, and New Zealand), but most were purchased from Platts McGraw-Hill (EU and UK) or NordPool Spot (Scandinavian countries). We use spot market prices where available. Otherwise, we use day-ahead prices. For many countries, separate peak and off-peak prices were available. In such cases we create a daily weighted average price, weighted by the fraction of hours in the peak- and off-peak periods. For countries whose wholesale price data were recorded at sub-daily intervals, we computed a quantity-weighted average price for the day. Finally, we use a similar method when multiple market nodes were available.9 All prices are

8

These prices are calculated as a weighted average of individual state REC prices, weighted by the RPS requirement in each state. States that do not have REC markets are excluded from this calculation. 9 This was the case, for example, for the NordPool countries. Platts country-level daily price data are already computed as spatially and temporally weighted averages.



11

Figure 2. Sample FIT, RPS, and REC data.

converted to constant 2010 US dollars.10 A detailed guide to our data sources is provided in the online appendix. When describing a stochastic economic outcome like wholesale electricity prices, variance is, in every practical sense, synonymous with risk. As such, our dependent variable is (log) quarterly variance in electricity prices at the country level. Denoting the wholesale electricity price as 𝑃+,- for country 𝑖 on day 𝑡, we compute 𝑉+,1 ≡ ln (𝑣𝑎𝑟[𝑃+,-∈1 ]), where 𝑠 indexes quarter. Figure 3 displays 𝑉+,1 for each country in our sample, and indicates over which interval of the 10

Conversions were made using monthly market exchange rates and annual GDP deflators for each country. Exchange rates are from the Federal Reserve Economic Data (FRED) database of the St. Louis Fed. GDP deflator data are from the World Bank Development Indicators database.



12

Figure 3. Log of quarterly variance in wholesale electricity price by country.

sample the country had either a FIT or RPS scheme in place. Table 2 presents summary statistics for our daily price data, our dependent variable, and quarterly control variables (explained below), as well as the fraction of observations (by country-quarter) for which FIT and RPS schemes were in place. Within our sample, FIT is clearly the dominant RES-E support policy, with 53 percent of country-quarters having an active FIT, in contrast to only 18 percent of country-quarters with an RPS. Moreover, FIT payments are substantially different across different renewable fuels, reflecting their differing costs. The average FIT payment for wind, one of the cheaper renewable fuels, is $0.05/KWh, whereas the average FIT payment for solar PV, one of the most expensive, is $0.17/KWh.



13

Table 2. Summary statistics. Daily variable Wholesale price (USD/MWh) Quarterly variables Variance, wholesale price Log(variance, wholesale price) FIT (any fuel) RPS Log(variance, cost per MWh natural gas fired) Log(variance, cost per MWh oil fired) Log(variance, daily maximum temperature within country) Log(population – millions) Log(GDP – trillions, 2014 USD) Log(CO2 emissions – billion metric tons) Total wind generation capacity (GW) Total solar PV generation capacity (GW) FIT payments by fuel (USD/KWh) Wind Solar PV Biomass Geothermal Ocean/tidal RPS requirement (percentage) Change in RPS requirement from same quarter, previous year

Mean 53.65

Std. Dev. 32.00

Min. -0.72

Max. 1,030.72

358.16 4.68 0.53 0.18 3.21

1,268.31 1.58 0.50 0.39 1.21

1.10 0.09 0 0 -0.33

25,424.79 10.14 1 1 5.83

3.76

1.26

-1.39

6.63

2.66

0.71

0.24

4.19

2.69 -0.53 -1.93

1.27 1.33 1.46

1.35 -2.96 -3.33

5.74 2.74 1.80

4.28 0.73

8.08 2.49

0.01 0

46.38 24.18

0.05 0.18 0.05 0.08 0.05 1.76 0.16

0.07 0.25 0.08 0.19 0.19 4.08 0.44

0 0 0 0 0 0 0

0.33 0.88 0.40 1.08 1.08 18.2 2.3

NOTE: (i) Total number of daily wholesale price observations is 46,857. (ii) Total number of quarterly observations is 520. (ii) The means for FIT and RPS are interpreted as the fraction of total quarterly observations for which each policy was in effect. (iii) All monetary values expressed in constant 2010 USD, unless otherwise noted.



4.2. Regression Model To test whether FIT or RPS policies are associated with a reduction in electricity price variance, we use ordinary least squares (OLS) to estimate the following regression equation: 𝑉+,1 = 𝛽 (>? 𝑅𝑃𝑆+,1 + 𝛽 ABC 𝐹𝐼𝑇+,1 + 𝜐𝑉+,1HI + 𝛿𝑋+,1 + 𝜂𝑍+,1 + 𝛼+ + 𝜃1 + 𝛾Q + 𝜀+,1 ,



14

(1)

where 𝑉+,1 is defined above. 𝑅𝑃𝑆+,1 and 𝐹𝐼𝑇+,1 are policy dummies, 𝛼+ are country fixed effects, 𝜃1 are quarterly fixed effects,11 and 𝛾Q are year fixed effects. We include a one-quarter lagged dependent variable term to control for potential autoregressive effects. 𝑋+,1 is a vector of RES-E capacities and policy characteristics. 𝑍+,1 a vector of additional quarterly control variables. Standard errors are clustered by country group-year, for a total of 51 clusters. This satisfies the rule of thumb of at least 50 clusters to avoid over-rejection in group-year panel data models (Cameron and Miller 2014). Country groups are chosen based on geographic adjacency and the clear similarities in countries’ time series for 𝑉+,1 (Figure 3). The NordPool countries comprise one country group, whereas the continental European Union countries form another. The U.S. and Canada are grouped together, but Australia, New Zealand, and the United Kingdom each comprise their own single-country groups. Our explanatory variables are as follows. First, to examine the effects of specific policy characteristics, because solar PV and wind generation are the two predominant RES-E technologies we include FIT payments to solar PV and wind. Additionally, greater RES-E capacity already in place should itself result in a stochastic merit-order effect, but would reduce the effect of either support policy. We thus include solar PV and wind generation capacities.12 We include the annual change in the RPS requirement to capture the incremental effect of increased stringency in RPS policy on new investment in RES-E generation. The intuition here is related to the idea of a binding versus non-binding RPS (Shrimali and Kniefel 2011).13 By

11

More accurately, 𝜃1 are seasonal fixed effects—we switch Q1-Q3 and Q2-Q4 for countries in the southern hemisphere. 12 Because the level of capacity investment is in part endogenously determined by the support policy, to circumvent any endogeneity issues we use lagged values here as well. 13 For example, say an RPS requirement of 10 percent is implemented in a country that already generates 12 percent of its supply from RES-E. The RPS would be non-binding, because no new RES-E would be needed to meet the requirement. Shrimali and Kniefel (2011) in fact find a negative effect of RPS on RES-E penetration at the US state



15

isolating the incremental RPS requirement, we are better able to observe the true incremental stochastic merit-order effect of the RPS, as opposed to the cumulative effect, which may in some cases include preexisting RES-E capacity. Finally, because natural gas (predominantly) and oil are typically used as the marginal fuels in electricity generation, the contemporaneous variation in their market prices should affect the variation in the wholesale price of electricity. We therefore compute the cost per MWh of natural gas- and oil-fired electricity at the prevailing daily spot prices using the relevant conversion factors, and calculate the (log) quarterly variances of each.14 We use daily Brent crude oil and Henry Hub natural gas spot prices, as we consider these to be reasonable global benchmark prices for each resource. We control for the effect of temperature-related demand shocks by including the (log) quarterly variance in country 𝑖’s daily maximum temperature.15 Potential socio-economic effects related to economy size and population are captured by log(GDP) and log(population). Finally, to control for possible effects related to other carbon emissions reduction policies (for example, emissions trading schemes), we include log(CO2 emissions). The primary threat to the validity of our estimates is the possibility that policy makers may choose RES-E support policies with the specific purpose of managing the volatility of electricity prices. If true, this policy endogeneity would likely bias our key coefficient estimates. In our view, however, cross-country differences in price variance are unlikely to be driving the variation in RES-E policy adoption. The stated justifications for most RES-E policies are to increase the amount of renewables in the electricity production system and to reduce both global level. This is because many states consider existing RES-E capacity as eligible under the policy, undermining the promotion of investment in new capacity. 14 The conversion factors are 10.1 Mcf per MWh for natural gas and 1.75 barrels per MWh for oil. Source: EIA http://www.eia.gov/tools/faqs/faq.cfm?id=667&t=2 15 Daily maximum temperature data are in degrees Celsius, and for each country are calculated as a national-level average across weather stations. Collected from NOAA (Menne et al. 2012a, b).



16

and local pollution from fossil fuel generation. Moreover, there are likely simpler and more effective policies to reduce electricity price volatility than FIT or RPS, such as implementing large-scale demand side management and energy efficiency programs. However, if policy makers do have price volatility in mind when choosing a RES-E support policy, then it should be that countries that eventually adopt such policies likely have higher electricity price variance than countries that do not. To test this hypothesis, we constructed a subsample of our data including only observations for which a given country did not have a RES-E policy in place. We then estimated an equation similar to (1), including an indicator variable that takes the value “1” if the country ever passed a RES-E policy, and “0” if the country never passed a RES-E policy. This provides a lens for us to examine whether countries that went on to pass RES-E policies were different in terms of electricity price variance than those without. We do not find any statistically significant evidence that the quarterly variance in electricity prices differed across these two sets of countries, alleviating the concern that the policies themselves were endogenously selected based on differences in price variance.16 The results of these regressions can be found in our Online Appendix (Table A1).

4.3. Estimation Results for Main Specification Table 3 presents our primary estimation results. We find a robust, statistically significant reduction in the variance of the wholesale electricity price under both FIT and RPS. We consider this to be convincing evidence of the relationship between RES-E support policies and the stochastic merit-order effect hypothesized in Section 3. 16

Our sample of countries is limited in this exercise, because some countries in our sample have a RES-E policy in

place over the entirety of our observation period.



17

Table 3. OLS estimates. Dependent variable: 𝑉+,1 ≡ log (𝑣𝑎𝑟[𝑃+,-∈1 ]). Variable RPS FIT (any fuel) Change in RPS requirement FIT payment to wind † FIT payment to solar PV † Total wind generation capacity † Total solar PV generation capacity † One-quarter lagged dependent variable Log(variance, cost per MWh natural gas fired) Log(variance, cost per MWh oil fired) Log(population)

(1) -0.28 (0.30) -0.52*** (0.17) -0.15 (0.17) 2.71 (2.42) -1.45*** (0.42) -0.01 (0.01) -0.10** (0.04) 0.13** (0.06) 0.28*** (0.11) 0.05 (0.11)

(2) -0.50*** (0.17) -0.21 (0.14) 2.41 (2.31) -1.43*** (0.42) -0.01 (0.01) -0.09** (0.04) 0.13** (0.06) 0.28*** (0.11) 0.05 (0.11)

2.52** (0.96) 446 0.61

2.56*** (0.95) 446 0.61

(3) -0.45* (0.25) -0.50*** (0.18) 2.53 (2.40) -1.44*** (0.42) -0.01 (0.01) -0.10** (0.04) 0.13** (0.06) 0.28*** (0.11) 0.05 (0.11)

Log(GDP) Log(CO2 emissions) Log(variance, daily maximum temperature) Constant

2.51** (0.97) 446 0.61

(4) -0.24 (0.28) -0.49*** (0.16) -0.18 (0.17) 2.06 (2.63) -1.30*** (0.46) -0.02 (0.01) -0.09** (0.04) 0.13** (0.06) 0.28*** (0.11) 0.04 (0.11) 4.07 (9.36) -0.94 (1.08) -0.34 (1.33) 0.04 (0.18) -8.57 (22.18) 446 0.62

(5) -0.47*** (0.16) -0.23* (0.14) 1.84 (2.54) -1.29*** (0.45) -0.02 (0.01) -0.09* (0.04) 0.13*** (0.06) 0.28*** (0.11) 0.04 (0.11) 4.36 (9.44) -0.99 (1.08) -0.26 (1.32) 0.04 (0.18) -9.01 (22.27) 446 0.62

(6) -0.45** (0.21) -0.48*** (0.17) 1.91 (2.63) -1.31*** (0.46) -0.02 (0.01) -0.09** (0.04) 0.14** (0.06) 0.28*** (0.11) 0.04 (0.11) 3.66 (9.37) -0.79 (1.07) -0.37 (1.32) 0.04 (0.18) -7.56 (22.24) 446 0.61

Number observations 𝑅U F-test of joint significance‡ F-statistic 3.58 4.93 p-value 0.06 0.03 NOTES: All models include country, year, and quarter fixed effects with robust standard errors clustered by country group-year (51 clusters). †Indicates lagged value (same quarter, previous year). ‡F-test of joint significance of RPS and Change in RPS requirement, where rejection of null indicates joint statistical significance. * p? , and 𝛾+ABC_(>? —i.e., estimates of the trends in (log) quarterly variance in electricity prices for each of the 24 country-policy intervals listed in Table 6. Figure 4 displays these



Note that 𝛽+q and 𝛾+q (𝑘 = 𝑁𝑂_𝑃𝑂𝐿, 𝐹𝐼𝑇, 𝑅𝑃𝑆, 𝐹𝐼𝑇_𝑅𝑃𝑆) are identified only when country 𝑖 has policy 𝑘 over some portion of our sample period. For instance, if a country has only FIT over the entire sample, this means only 𝛽+ABC and 𝛾+ABC are identified in (3). 21



27

Figure 4. Estimates of trend in log (𝑣𝑎𝑟[𝑃+,-∈1 ]) for each country-policy interval (see Table 6).

estimates. As a benchmark, we show the mean estimated trend across the five country-policy intervals with no RES-E support policy. From Figure 4 it is clear that countries with a FIT generally seem to experience a decline over time in the quarterly variance in electricity prices, whereas for countries with either an RPS or both policies combined the results are mixed. The second stage of our long-run analysis is thus to regress the estimated trends on a set of policy indicator variables: 𝚪 = 𝜑 mn_>np + 𝜑 ABC 𝐹𝐼𝑇+ + 𝜑 (>? 𝑅𝑃𝑆+ + 𝜑 ABC_(>? 𝐹𝐼𝑇_𝑅𝑃𝑆+ + 𝜇+ ,

(4)

where 𝚪 is a column vector of the estimates 𝛾+mn_>np , 𝛾+ABC , 𝛾+(>? , and 𝛾+ABC_(>? for each of the 24 country-policy intervals in Table 7. Regression equation (4) represents a simple 𝑡-test of means.



28

Table 8. Estimated effects of RES-E policies on the long-run trend in log (𝑣𝑎𝑟[𝑃+,-∈1 ]). FIT RPS FIT and RPS No policy

(1) -0.12*** (0.02) 0.07 (0.09) 0.00 (0.14) 0.08*** (0.01)

(2) -0.09*** (0.02) 0.28 (0.35) 0.15 (0.22) 0.07*** (0.01)

(3) -0.08*** (0.03) -0.01 (0.02) -0.37 (0.38) 0.07*** (0.01)

Control variables in first-stage Log(variance, cost per X X MWh natural gas fired) Seasonal fixed effects X Log(variance, cost per MWh oil fired) Log(GDP) Log(variance, daily maximum temperature) Observations 24 24 22 0.35 0.30 0.31 𝑅U Efron standard errors reported. * p