ï»¿ WPS6376
Policy Research Working Paper 6376
How Fit are Feed-in Tariff Policies?
Evidence from the European Wind Market
Fan Zhang
The World Bank
Europe and Central Asia Region
Office of the Chief Economist
February 2013
Policy Research Working Paper 6376
Abstract
Feed-in tariffs have become the most widely used policy yield greater levels of wind installation. Non-economic
instrument to promote renewable energy deployment barriers and rent-seeking may have contributed to the
around the world. This paper examines the relation weak correlation. On the other hand, the length of feed-
between tariff setting and policy outcome based on in contract and guaranteed grid access are important
wind capacity expansion in 35 European countries determinants of policy effectiveness. A one-year extension
over the 1991â€“2010 period. Using a dynamic panel of an original 5-year agreement on average increases
data model, it estimates the long-run elasticity of wind wind investment by 6 percent annually, while providing
deployment with respect to the level of feed-in support. an interconnection guarantee almost doubles wind
The analysis finds that higher subsidies do not necessarily investment in one year.
This paper is a product of the Office of the Chief Economist, Europe and Central Asia Region. It is part of a larger effort by
the World Bank to provide open access to its research and make a contribution to development policy discussions around
the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author may be
contacted at fzhang1@worldbank.org.
The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.
Produced by the Research Support Team
How Fit are Feed-in Tariï¬€ Policies? Evidence from the
European Wind Market
âˆ—
Fan Zhang
Key Words: Feed-in Tariï¬€s, Wind Energy, Europe
JEL Classiï¬?cation: L52, L98, Q28, Q48
Sector Board: Energy and Mining
âˆ—
World Bank Energy Department of Europe and Central Asia Region. Email: fzhang1@worldbank.org.
I thank Uwe Deichmann, Carolyn Fischer, Joshua Linn, Govinda Timilsina and participants at the World
Bank SDN BBL seminar for helpful comments and suggestions. I thank Louis Preonas and Xu Chen for
excellent research assistance. Financial support from the World Bank Research Support Budget is gratefully
acknowledged. All errors remain those of the author.
1
1 Introduction
Governments have implemented various incentive programs to accelerate renewable energy
deployment around the world. In the 1970s and 1980s, the policy emphasis was to support
technology research and development. In the 1990s, the focus gradually shifted to stimulate
market demand for renewable energy. The two most commonly used demand-side policies
are price-based feed-in tariï¬€s (FiTs) and quantity-based quota obligations (also referred to
as renewable portfolio standards [RPS]). Under a FiT scheme, governments set prices often
at a premium for diï¬€erent types of renewable power to compensate producers for the higher
cost of producing clean energy. Utilities are required to purchase power from renewable
resources at this price, but can either spread the additional costs across their entire cus-
tomer base or receive compensation from the government to recover the incremental costs.
Essentially, FiTs work as subsidies to renewable energy to make it cost competitive to fossil
fuel based technologies. In contrast, a quota obligation creates a market for tradable green
certiï¬?cates (TGCs), which are awarded to renewable producers based on their renewable
energy output. Electricity suppliers must purchase certiï¬?cates or otherwise supply renew-
able energy for a certain percentage of their total end-use deliveries. The market value of
a TGC thus reï¬‚ects the balance between the supply of renewable energy and the quantity
demanded by the regulation.
Since ï¬?rst introduced in Germany in 1991, FiT schemes have gained greater popularity
worldwide. As of early 2012, at least 65 countries have enacted FiT policies, while 18
countries opted for RPS (REN21, 2012). FiTs have for the most part been viewed as suc-
cessful in terms of deployment: by some estimates, they are responsible for approximately
75 percent of global solar photovoltaic and 45 percent of global wind capacity. However,
their record for cost-eï¬€ectiveness is more inconsistent1 . In particular, setting the â€?rightâ€?
1
In Spain, for example, an overly generous solar FiT resulted in $23 billion subsidy payments and huge
power plant-sized installation of photovoltaics. The government had to later reduce the tariï¬€ by 45 percent
for new solar plants amid budget constraints. Similarly, in Ukraine, the government has introduced a solar
FiT at 46 euro cents/kWh, the highest in Europe and more than twice Germanyâ€™s in 2012. This has
prompted concern that tariï¬€s are set higher than necessary and may provide excessive windfall proï¬?ts for
2
level of support is diï¬ƒcult. On the one hand, a FiT has to provide suï¬ƒcient incentives to
achieve an overall quantity of renewable generation; on the other hand, it should not be too
generous to allow poor performing investment to survive solely based on heavy subsidies.
Furthermore, if renewable subsidies are paid by energy consumers, high FiTs may impose
an adverse impact on growth and aï¬€ordability. This is a particular concern to developing
countries, where households devote a larger proportion of income to energy and are more
vulnerable to rising tariï¬€s.
To understand the cost-eï¬€ectiveness of existing FiT policies, this paper empirically
assesses the responsiveness of renewable investments to various levels of FiT support based
on onshore wind development in Europe. We also explicitly account for other policy design
features and electricity market conditions that may impose a signiï¬?cant impact on the
eï¬€ectiveness of FiTs. Empirical studies on the cost-eï¬€ectiveness of FiT schemes have
been sparse, partly due to the complexity of the policy design and the lack of comparison
data across countries. In our data collection eï¬€ort, we attempt to rectify the problems by
reviewing the detailed design features of wind FiTs in 46 European countries over the period
1991-20102 . Because FiT rates often vary by size and location3 , we obtain generator-level
wind installation outcomes in order to match wind investment and FiT polices as closely
as possible.
Speciï¬?cally, we estimate a dynamic panel data model to account for (1) autoregres-
sive investment cost shocks, (2) potential correlation among FiT rates, wind installation
and unobservable investment costs, and (3) plausible sequential exogeneity of FiT rates -
past investment performance aï¬€ects the current subsidy level. We ï¬?nd no evidence that
higher subsidies encourage higher uptake of wind power. The disconnect could indicate
the presence of non-economic barriers such as administrative hurdles that worked against
ï¬?nancial incentives. It could also suggest a correlation between remuneration and invest-
investors.
2
Among the 46 countries, 35 have either adopted FiTs or RPS by 2010, while 11 countries have never
enacted any renewable incentive policies during the observation period.
3
For instance, Greece oï¬€ers separate rates for mainland vs.island wind installations. Many countries
restrict FiTs to installations below a certain size (such as below 20 MW), or have size-speciï¬?c rates.
3
ment costs - high remuneration allowed deployment in high-cost locations (with poor wind
conditions) or encouraged rent-seeking behavior 4 - both could increase investment costs
and reduce the average investment elasticities 5 . After controlling for the cumulative wind
electricity generating capacity, we ï¬?nd FiT price has a small but negative correlation with
wind generation. Because turbine design and the selection of wind location are important
determinants of generation eï¬ƒciency (output per unit of installed capacity), this result
suggests that high subsidies provide fewer incentives for investors to avoid high cost sites 6
or engage in innovation activities to enhance turbine eï¬ƒciency.
In contrast, guaranteed grid access and the duration of FiT contracts are crucial policy
characteristics inï¬‚uencing the impact of wind support. Moving from no guaranteed grid
access to guaranteed grid access can almost double wind installation in one year, ceteris
paribus ; extending the contract length by an additional one year of an original 5-year
agreement will on average increase wind investment by 6 percent annually. The results
conï¬?rm that investors attach high premiums to the certainty of investment return. A
predictable long-term policy commitment is likely to be more eï¬€ective than excessive short-
run ï¬?scal incentives to attract investment.
The organization of the power market also has a sizable impact on the deployment of
wind power. If FiT rates are exogenous to electricity prices, higher electricity prices reduce
the relative price premium of FiTs, hence the amount of wind investment. But when
electricity is traded in a competitive wholesale market, higher electricity prices appear to
strengthen the incentive for renewable energy development. This is because many countries
allow renewable generators to either sell their electricity at the guaranteed FiTs or sell it
in the open exchange. As a result, wholesale competition allows renewable developers to
beneï¬?t from higher market prices especially during the peak demand.7 We ï¬?nd that with
4
An IEA(2011) study suggests that solar panel sellers price their systems according to the incentive in
a given country and try to take a share of any excess remuneration.
5
An incidence analysis on who beneï¬?ts from high FiT rates could be an interesting topic for future
research.
6
Good wind locations are unlikely to be exhausted in most countries in the sample as existing wind
generation only represents a small percentage of the estimated wind economic potential in these countries.
7
In Turkey, average electricity price in the wholesale market was higher than FiTs before 2010. As a
4
the presence of a competitive wholesale market, a one standard deviation increase in an
electricity price (3.79 euro cents/kWh) is on average associated with a 38-70 MW increase
in wind installation, all other things being equal.
The remainder of the paper proceeds as follows. Section 2 provides a brief review of the
existing empirical literature on the eï¬€ectiveness of renewable incentive policies. Section 3
develops a ï¬?rm model of capital investment, and identiï¬?es the empirical speciï¬?cation and
the estimating strategy. Section 4 describes the data. Section 5 presents the results and
Section 6 concludes.
2 Literature Review
Empirical analysis on the cost-eï¬€ectiveness of FiTs has been sparse. A few relevant studies
are summarized in this section. The International Energy Agency (IEA) (2008) chooses
as an indicator the ratio of annual additional installation of renewable energy and the
remaining renewable energy potential to assess the eï¬€ectiveness of renewable policies in 35
OECD and BRICS countries over the period 2000 to 2005. Comparing the eï¬€ectiveness
indicator against the level of FiTs, the study ï¬?nds that â€œbeyond some minimum threshold
level of 0.07 USD/kWh, higher remuneration levels do not appear to yield greater levels of
policy eï¬€ectivenessâ€?. The study is not based on an econometric analysis. It also has not
considered the impact of other policy design features, such as mandate grid access and the
longevity of purchase contract.
Building upon a similar methodology, IEA (2011) expands the analysis to evaluate the
average impact of FiT policies between 2001 and 2009. With the extended observation
period, the report ï¬?nds an inverse correlation between the level of support and the average
impact for FiTs (but the result is not statistically signiï¬?cant). The report suggests that
the negative correlation may be explained by a learning eï¬€ect: countries where markets are
functioning well and have become mature tend to see both high deployment and low remu-
result, most of the renewable projects had chosen to sell into the wholesale market. Zhang(2009) explains
how electricity market reform has helped the market penetration of renewable energy in Turkey.
5
nerations. However, the report also admits that rigorous analysis regarding the temporal,
dynamic dimension of impact and remuneration adequacy would be required to gain a full
understanding of the direction of the causality.
Mulder (2008) analyzes the performance of feed-in systems in stimulating wind power
investment in EU15 countries between 1985 and 2005. Mulder ï¬?nds that Germany, Den-
mark, and Spain are consistently ranked the top performers measured by annual investment
growth rate, total installed capacity, and realizing the full potential of wind resources. All
four countries primarily used FiT systems to encourage wind deployment. Since FiTs have
never been very high in Germany or Spain, the study concludes that economic incentive is a
necessary but not a suï¬ƒcient condition in achieving widespread renewable energy adoption.
Other factors such as a long-term and stable policy environment may be more important
determinants.
oderholm and Klaassen (2007) provide a quantitative analysis of innovation and dif-
SÂ¨
fusion in the wind power sector in Denmark, Germany, Spain and the United Kingdom
during 1986-2000. Their empirical analysis estimates a simultaneous relationship between
FiT rates and the investment cost of wind turbines. On the one hand, higher feed-in prices
promote the diï¬€usion of wind technology which in turn encourages learning and cost reduc-
tion. On the other hand, high subsidies for wind are associated with lower learning rates
- lower cost reduction per doubling of installed capacity. Overall, the authors ï¬?nd the net
eï¬€ect of FiT rates on wind cost ambiguous.
Dong (2012) examines the relative eï¬€ectiveness of FiT and RPS in promoting wind
technology in 53 countries from 2005 to 2009. Consistent with individual country case
Ä±o and Gual[2007]), the
studies (Astrand and Neij[2006], Butler and Neuhoï¬€[2005] and RÂ´
study identiï¬?es a strong correlation between the implementation of FiT policies and de-
velopment of the wind market. But the analysis treats all FiT policies the same (using
a policy dummy variable), it does not examine the impact of heterogenous policy design,
including the size of FiT rates.
Overall, no study has systematically examined the cost-eï¬€ectiveness of existing FiT
schemes. In this paper, we intend to ï¬?ll the gap in the literature by estimating the elasticity
6
of renewable investment to FiT subsidies, while also taking into account other nuanced
policy characteristics. We employ a dynamic panel data model and use generalized methods
of moments (GMM) to address empirical challenges related to the dynamic correlation
among wind installation, investment costs, and FiT remuneration.
3 Model
3.1 Investment Theory
In this section, we develop a simple ï¬?rm model of capital investment to motivate the
empirical speciï¬?cation. Suppose in each period t, the ï¬?rm chooses the optimal amount
of capital Kt (megawatts[MW]) to maximize the expected present value of net cash ï¬‚ows
from the invested wind capacity:
N
max Vt = Et Î² t (pt Qt ) âˆ’ gt Kt (1)
t=0
s.t. Qt = At Kt
where Et is the expectation sign. Î² = 1/(1 + r) is the discount factor, and r is the
discount rate. Qt is the amount of generation and is assumed to be subject to a Cobb-
Douglas production function. Wind generation has zero fuel cost and very low operation
and maintenance costs (represent about 2.5 percent of the total costs). Almost 75-80
percent of wind production cost is related to upfront investment, including capital expenses
on wind turbine, foundations, electrical equipment, grid connection and so on (EWEA,
2012). Therefore, only capital input is included in the production function8 . ( < 1) is
the elasticity of the wind supply curve. Assuming wind resources vary across regions, total
wind generation falls oï¬€ less than linearly with total wind capacity. At is the load factor
(the average energy output as a percentage of total theoretical capacity) and is mostly
determined by the availability of wind (wind speed) and the technical eï¬ƒciency of wind
turbines.
8
We also ignore ï¬?nancing cost, which is estimated to be about 1.2 percent of the total cost (EWEA,2012).
In section 5, we conduct robustness check by including labor costs in the estimation model.
7
pt is the price at which the ï¬?rm will be paid for wind electricity generation. When FiT
policy is in operation, pt is the guaranteed FiT rate (ft ). When FiT expires, electricity is
sold at current market price (et ). Therefore, Et (ps ) = ft if s â‰¤ N and Et (ps ) = Et (es ) = et ,
if s > N , where N is the length of the guaranteed feed-in contract.9 gt is the unit capital
investment cost in period t (euro/MW).
Solving the above proï¬?t maximization problem, the optimal amount of investment in
period t is given by:
N t
t=0 Î² Î±E (pt )
1
âˆ—
Kt =( ) 1âˆ’ (2)
gt /At
Taking natural log on both sides of Equation (2) gives:
âˆ— 1
logKt = (logBt + logEt (pt ) âˆ’ log (gt /At )) (3)
1âˆ’
N t
where Bt = ( t=0 Î² )/ .
By aggregating the above equation across ï¬?rms, one obtains an industry demand func-
tion for capital, which is a function of the interest rate (r), the duration of FiT contract(N ),
the expected remuneration of wind generation (Et (pt )), and the unit investment cost
(gt /At ). Due to the correlation with load factor, the unit investment cost is now mea-
sured in euro/megawatt hours and is site speciï¬?c. The lower the wind speed, the higher
the cost10 .
Equation (3) implies that ï¬?rm investment on wind capital is positively correlated with
FiT rate through the expected value of the marginal unit of new capital (Et (pt )). But the
positive correlation could be diminished if high FiT rates induce rent-seeking of system
suppliers so that any excessive compensation is captured by system suppliers (increase in
gt ) or pushing investment to high cost locations (decrease in At ).
9
We assume that electricity prices follow ï¬?rst-order Markov processes and their conditional expectations
are given by: Et (es ) = et .
10
The local wind climate is the most important factor in determining wind investment cost. For example,
it is estimated that the cost of wind electricity ranges from approximately 7-10 euro cents/kWh at sites
with low average wind speeds to approximately 5-6.5 euro cents/kWh at a wind site with average speeds
(EWEA, 2012).
8
3.2 Basic Econometric Model
Based on Equation (3), a basic empirical model can be described as the following:
logKit = Î²1 logfit + Î²2 logeit + Xit Î³ + Tt + zt + (vi + cit + uit ) (4)
cit = Î±ci,tâˆ’1 + it |Î±| < 1 (5)
where logKit is the log annual wind capacity additions measured in MW in year t in
country i. logfit is the log FiT rate measured in euro cents/kWh. It is the key variable of
interest. The coeï¬ƒcient of logfit measures the elasticity of investment with respect to FiT
incentives. As shown above, the degree of sensitivity of investment to FiT prices depends
on the correlation between FiT and investment costs. A positive correlation between fit
and gt /At due to choices of high cost locations or rent-seeking implies low sensitivity (a
small/zero coeï¬ƒcient of logfit ). The extent of non-economic barriers (such as planning
delays and restrictions, lack of co-ordination between diï¬€erent authorities, long lead times
in obtaining authorizations etc. (IEA, 2008)) also aï¬€ects the size of the coeï¬ƒcient. If
investors in countries with high FiT rates also happen to face higher costs associated with
dealing with non-economic barriers, the investment elasticity will be smaller.
logeit is the log of the average electricity price. Ideally, we would want to use annual
average wholesale prices. But because wholesale prices are only available for a small set of
countries in the sample (IEA, 2012), we use average end-user prices as a proxy.11 Higher
electricity price could encourage renewable deployment if investors take it as a sign of future
higher energy demand. On the other hand, if FiT rates are independent from electricity
prices, higher electricity prices make current investment in renewable energy less attractive
because the relative price premium for renewable electricity becomes lower. There is no
prior on the sign of Î²2 .
Xit is a vector of other time-varying variables aï¬€ecting ï¬?rm investment decisions. They
include (1) Nit , the log of the length of guaranteed FiT contract; (2) Git , a dummy variable
11
In the robustness check, we restrict the sample to countries where wholesale electricity prices are
available and re-estimate the model based on average wholesale prices. We ï¬?nd the basic conclusions do
not change with the alternative data.
9
equals one if there is an interconnection guarantee (both technical requirements and legal
procedures) to renewable electricity. The longer the guaranteed FiT support, the lower the
uncertainty of investment return over the lifetime of the project. The coeï¬ƒcient of Nit is
expected to be positive. Similarly, guaranteed grid access reduces investment risk, and the
coeï¬ƒcient associated with grid access is likely to be positive.
(3) Cit is a dummy variable equal to 1 if a country has either partially or fully developed
a competitive electricity wholesale market, and 0 otherwise. The regulatory regime could
aï¬€ect clean energy investment, but the direction of the impact is ambiguous (Heiman, 2006).
On the one hand, market competition ensures consumer choice and promotes product dif-
ferentiation. Consumer demand for â€œgreenâ€? energy may boost renewable energy supply. A
competitive market is also more friendly to small-scale decentralized power generation as
compared to traditional rate-of-return regulation that favors the capital-intensive invest-
ment of large central power plants. On the other hand, renewable technologies in general
are still more expensive than conventional generation12 . Without policy support, opening
electricity markets to competitive forces might result in less renewable energy development
as electricity providers turn to the cheapest source of electricity.
(4) Sit is the log of total power output (measured in terawatthours [TWh]) and is an
indicative proxy for demand shocks of both domestic and export consumption. The higher
the demand for electricity, the more electricity generating capacity including that from
renewable sources is likely to be built. The coeï¬ƒcient associated with Sit is expected to be
positive.
The rapid global development of wind power has had a strong inï¬‚uence on the cost
of wind energy over the last 20 years. Tt is a time variable used to capture the common
technology trend towards lower costs of the wind industry. We also include year dummies zt
representing year-speciï¬?c intercepts to account for other common cyclical components. For
a dynamic panel data estimation, including common time eï¬€ects helps reduce the eï¬ƒciency
loss arising from cross section dependence (Phillips and Sul, 2003).
12
Especially considering the extra costs on infrastructure required to integrate distributed and intermittent
renewable generation.
10
vi is a country ï¬?xed-eï¬€ects variable reï¬‚ecting any time-invariant diï¬€erences across coun-
tries. This may include the natural endowment of wind resources, the initial land avail-
ability and public acceptance of wind technology, and any other national renewable energy
policies consistently enacted throughout the sample period.
cit is the unit investment cost. Investment costs diï¬€er signiï¬?cantly across countries
(IEA, 2011) but the cost data are generally not available at a disaggregated level for all
countries. Further, investment costs are likely to be serially correlated, as described in
Equation (5) where Îµit is the stochastic term.
Finally, uit are serially uncorrelated shocks. The error components are in parentheses.
Coeï¬ƒcients Î²1 , Î²2 , Î³ and Î± are parameters to be estimated.
In addition to stimulating investment on wind capacity, a more direct purpose of FiT
is to increase actual output of renewable electricity. To assess the direct impact of FiT on
wind generation given existing wind capacity, we estimate the following model:
Â¯ it + Î»2 logfit + Î»3 logeit + Xit Ï‘ + Tit + zt + (vi + uit )
logGit = Î»1 log K (6)
where Git is the total amount of wind electricity generated (in TWh) in country i for year
Â¯ it is the cumulative wind capacity. All other variables are deï¬?ned in the same way as
t. K
perviously. Î»1 , Î»2 , Î»3 , and Ï‘ are parameters to be estimated.
Once a wind project is developed, one would expect investors to maximize production
as much as possible because the variable cost is low. Assuming grid access is obtained, the
capacity factor of a wind power plant is only determined by wind speed, altitude and turbine
eï¬ƒciency (EWEA, 2012). The hypothesis we want to test is whether higher FiTs aï¬€ect
generation eï¬ƒciency (output per unit of capacity installed). If generous subsidies allow
investment in lousy wind locations to survive, we will expect the coeï¬ƒcient associated with
log of FiT rates to be negative. On the other hand, if FiT policies provide incentives for
innovation activities aimed at increasing turbine eï¬ƒciency, the coeï¬ƒcient of log of FiT rates
will be positive.
It should be noted that if several countries trade electricity on a common exchange, then
renewable incentives in neighboring countries will also aï¬€ect investment and production in
11
the domestic market. The current models do not capture such a spillover eï¬€ect.
3.3 Estimating Strategy
In the following, we discuss estimating strategies based on our assumptions of the error
term. The number of countries in our unbalanced panel data is 35 (N ), and the number
of years (T ) per country ranges from 1 to 20 with an average of 6.4 years. So we are
interested in consistent estimation of the parameters under the premise of small T and
large N . If the country-speciï¬?c eï¬€ects vi are correlated with other explanatory variables,
which in turn are orthogonal to error components cit + uit , then OLS estimation would
be biased but a within transformation will yield consistent estimators. However, if the
explanatory variables are also correlated with other error components in addition to the
correlated country ï¬?xed eï¬€ects, then we would need instrumental variables to purge the
endogenous component.
In fact, we suspect the FiT rates are likely to be correlated with the unobserved in-
oderholm and Klaassen (2007) and IEA(2011), higher FiT
vestment costs. As noted in SÂ¨
remuneration encourages installation in high-cost sites with poor wind resources. In addi-
tion, higher subsidies may push up system prices as suppliers along the value chain try to
extract excess proï¬?ts. In both cases, higher FiT rates are correlated with higher investment
costs. Higher investment costs would in turn inhibit the ability of the industry to expand
wind capacity. Then under the ï¬?xed T and large N assumption, both OLS and ï¬?xed-eï¬€ects
models yield biased estimation of Î²1 13 . Speciï¬?cally, if E (fit cit ) > 0 and E (Kit cit ) < 0, a
within-estimator would be downward biased.14
Due to the autoregressive structure of the investment costs (cit ) in the error term, we
cannot use lagged values of FiT price in levels as instruments in an equation expressed
in ï¬?rst-diï¬€erence, or lagged ï¬?rst-diï¬€erences of endogenous variables as instruments for an
equation in levels. Blundell and Bond (2000) shows that in a model where explanatory
variables are correlated with both ï¬?xed eï¬€ects (vi ) and the error term (cit ), and the error
13
If T is large, while N is small, then a ï¬?xed-eï¬€ect estimator would be asymptotically unbiased.
14
The direction of bias of an OLS model also depends on the sign of correlation between fit and vi .
12
term is autocorrelated, one can employ a dynamic (common factor) representation which
involves Î±-diï¬€erencing Equation(4) to partial out the unobserved time-varying variable15 :
logKit =Î²1 logfit âˆ’ Î±Î²1 logfi,tâˆ’1 + Î²2 logeit âˆ’ Î±Î²2 logei,tâˆ’1 + Xit Î³ âˆ’ Xi,tâˆ’1 Î±Î³
(7)
+ Î±logKi,tâˆ’1 + (zt âˆ’ Î±zi,tâˆ’1 ) + (1 âˆ’ Î±)vi + it + uit âˆ’ Î±ui,tâˆ’1
which can be rewritten as:
logKit =Ï?1 logfit + Ï?2 logfi,tâˆ’1 + Ï?3 logeit + Ï?4 logei,tâˆ’1 + Xit Ï?5 + Xi,tâˆ’1 Ï?6
(8)
âˆ— âˆ—
+ Î±logKi,tâˆ’1 + zt + vi + wit
The common factor restrictions are:
Ï?2 = âˆ’Î±Ï?1 , Ï?4 = âˆ’Î±Ï?3 and Ï?6 = âˆ’Î±Ï?5 (9)
If the assumption Î± = 0 holds, then Equation (8) can be also interpreted as that the wind
capacity expansion exhibits state dependence: the current level of investment depends on
investment in last period even after controlling for other variables included in the model.
In the absence of measurement error, the error term wit in Equation (8) is no longer
serially correlated after the transformation of the model. One can use diï¬€erence GMM
technique outlined in Arellano and Bond (1991), or system GMM proposed by Arellano
and Bover (1995) and Blundell and Bond(1998) to obtain consistent estimators of Equation
(8).16 On the other hand, as we donâ€™t expect serially correlated error term in Equation (6),
we estimate the determinants of wind generation using a ï¬?xed eï¬€ects model.
One concern arising from the use of GMM estimation of dynamic model (8) is that FiT
rates may be sequentially determined, that is there could be feedback from past renewable
investment to current values of FiT support. For example, FiTs can be revised downwards
when capacity expended and costs of installation reduced. If this is the case, then even
though there may not be contemporaneous correlation between wit and fit , fit may still be
correlated with wi,tâˆ’1 in a ï¬?rst-diï¬€erenced model, i.e. E ( logfit wit ) = 0. We therefore
include fi,tâˆ’1 and earlier lags as instruments for FiT rates in period t.
15
The above equation is obtained by subtracting Î±logKi,tâˆ’1 from both sides of Equation(4).
16
If there are measurement errors and wit follows a M A(1) process, we use Kis (s â‰¥ 3) as instruments in
the ï¬?rst-diï¬€erenced equations and Ki,tâˆ’s (s = 3) as instruments in the levels equations.
13
Another potential concern is the endogeneity of electricity price in both static and
dynamic equations (6) and (8). Wind power could aï¬€ect electricity prices through two
channels depending on the market structure. First, because wind power normally has a
low marginal cost and enters near the bottom of the supply curve, it shifts the supply
curve to the right. In general, power prices will be lower during periods with high wind,
the so called â€?merit-order eï¬€ectâ€?. In countries where wind penetration is high, the eï¬€ect
can have a large impact on wholesale prices so that electricity price is correlated with wind
installation and generation contemporaneously17 . Second, using more wind power increases
the average costs of production. If FiT subsidies and the costs of interconnecting a wind
project to the electric grid are paid by the rate payers, then higher wind generation would
lead to higher electricity tariï¬€s.
To deal with the endogeneity of electricity prices, we use the lagged values of ei,tâˆ’s (s â‰¥
2) as the instruments in the diï¬€erence equation and (ei,t âˆ’ ei,tâˆ’1 ) as the instrument in the
levels equation. We also augment the standard GMM estimation with an external instru-
ment - the average industrial natural gas prices. Gas prices constitute a relatively large
share of the total cost of gas ï¬?red power generation and gas ï¬?red power plants tend to be
the marginal producer of electricity that sets the market price. The additional instrumen-
tal variable from outside the structural equation introduces some exogenous variations in
the electricity price, and helps avoid the weak instrument problem that often arises in the
GMM estimation (Blundell and Bond (2007) and Bond (2002)). We assess the validity of
the external instrument and those from the GMM procedures using standard test of the
overidentifying restrictions.
17
For example, the reduction in wholesale price due to wind power in Ireland for 2011 is projected to
match the premiums that are paid to wind power generators, i.e. wind power support in Ireland is cost
neutral(Cliï¬€ord and Clancy, 2011).
14
4 Data
We employ an unbalanced panel of 35 European countries where annual wind installa-
tion and incentive policies for wind investment are observed during 1991-2010.18 The
information on renewable policy design was obtained mostly from IEAâ€™s renewable policy
database, and cross-checked with government websites, legislative texts and other related
publications. Our review indicates a great deal of variation and complexity over the design
of FiT policies19 . First, FiT rates can be deï¬?ned very speciï¬?cally according to the instal-
lation location and size. For instance, Greece oï¬€ers separate rates for mainland vs.island
wind installations. Many countries restrict FiTs to installations below a certain size (such
as below 20MW), or have size-speciï¬?c rates. Given the nonuniformity of FiT rates, we
collect generator-level wind installation data from a global wind farm dataset20 to match
wind installation with speciï¬?c policies.
Second, policies sometimes have built-in digression rates so that FiT rates for existing
contracts decrease over time. Four countries in our sample, i.e. Austria, France, Germany
and Switzerland, oï¬€er stepped reductions in tariï¬€ for existing contracts but not until at
least after 10 years in the contract. For instance, in France, wind tariï¬€s will be reduced
by 2 percent annually starting from year 11. We did not speciï¬?cally estimate the eï¬€ect
of rate digression on current investment, recognizing that ignoring future rate reduction
may inï¬‚ate actual FiT remuneration. However, since price adjustment occurs far into the
future, the present value of the price reduction is likely to be small. We also exclude these
four countries from the sample for robustness check. The main results do not change.
Third, although in most cases FiT rates are ï¬?xed over the contract length (standard
FiT), rates can also vary annually according to electricity market price changes or other
factors (variable FiT). For example, the ï¬?rst type of FiT enacted in Germany in 1991
involves an energy regulator releasing a new FiT rate each year. Several countries oï¬€er a
18
The 35 countries include 23 European Union member states, Switzerland, Norway, and 10 countries in
Eastern Europe.
19
See Fischer and Preonas (2012) for a thorough review of the various design of FiT policies.
20
http://www.thewindpower.net/
15
ï¬?xed premium on top of a (variable) electricity price (premium FiT). They can sometimes
be guaranteed for a ï¬?xed number of years (as in Denmark, Norway, and Netherlands) or
come with no such guarantee (in which case the length of contract will be 0).21 Because
wholesale electricity prices are generally not observed, the ï¬?nal remuneration price under
a premium FiT is unobtainable. In the empirical analysis, we restrict samples to standard
and variable FiTs, and introduce policy dummy variables to diï¬€erentiate these two types
of policy design.
Finally, four countries in the sample, i.e. Poland, UK, Italy and Sweden opted for a
quota obligation scheme facilitated with TGCs to achieve renewable targets22 . There is a
rich literature on the comparison of policy eï¬€ectiveness of FiT and TGC. Although this is
not the focus of this paper, we nonetheless collect data on the annual average certiï¬?cate
prices and estimate the correlation between the value of TGCs and the wind installation
results.
It should be noted that in addition to FiTs and TGCs, many countries have enacted
additional renewable energy provisions, such as investment tax credits, grants, or VAT
exemptions. State and local renewable policies are also prevalent (Fischer and Preonas,
2011). Country ï¬?xed-eï¬€ects will reï¬‚ect these additional incentives if they are stable across
years. However, if additional incentive policies are adjusted according to FiT remuneration
levels, then our estimated investment elasticity with respect to the FiT rates will be biased.
Data on annual wind generation are obtained from IEA world energy statistics and
balances database. All price variables are deï¬‚ated by the CPI and are expressed in 2000
prices.
21
Spain, Slovenia, and the Czech Republic allow generators to pick between the standard and premium
FiTs. In the dataset, the standard FiTs are assumed to be the relevant ones, because these represent the
more guaranteed incentive. No deï¬?nitive reference was found as to which policy wind generators are likely
alez [2008]) implies that ï¬?xed FiTs were the preferred
to pick, but a 2008 review of Spanish FiTs (GonzÂ´
incentive scheme.
22
Italy has implemented both FiT and TGC since 2008. But no wind installation was observed under
FiT. In addition to TGC, UK started implementing a premium FiT in 2010.
16
4.1 Descriptive Analysis
Table (1) presents summary statistics. The rightmost column reports simple AR(1) spec-
iï¬?cation for the series of variables based on system GMM estimates. All these series are
found to be highly persistent, although none appears to have an exact unit root23 . Figure
(1) shows that the number of countries applying FiT policies increased from ï¬?ve in 1991 to
31 in 2010. In comparison, fewer countries opted for RPS/TGC mechanism with just ï¬?ve
in 2010.
We divide the sample into diï¬€erent groups based on country-year observation on the
status of renewable policy. Table (2) reports the mean value of country characteristics
by group. The ï¬?rst three columns of Table (2) reveal statistically signiï¬?cant diï¬€erences
in country characteristics and policy results. Countries with lower wind potential, lower
electricity demand, lower electricity and natural gas prices, and no competitive wholesale
market are less likely to enact any renewable incentive policies. These countries have seen
almost no wind installation. The last three columns of Table (2) further compare the
three subcategories of FiT policies. Standard FiT oï¬€ers the longest contract year. It is
also associated with the highest average annual wind installation and generation. These
aggregate statistics suggest that (1) renewable policies seem to be eï¬€ective in stimulating
clean energy uptake; (2) multiple factors aï¬€ect both policy adoption and policy outcome,
such as renewable resource endowment, and electricity demand and prices24 ; (3) policy
certainty is an important factor aï¬€ecting the results of renewable development. Standard
FiT, Variable FiT and TGC market oï¬€er similar levels of annual remuneration to wind
projects, but since standard FiT provides the highest degree of certainty over the return
to an investment, it is associated with the largest amount of renewable deployment25 .
23
A standard diï¬€erence-GMM estimator will collapse if there is a unit root.
24
The correlation between market competition and the adoption of renewable policies may simply indicate
that the general business environment is more dynamic and receptive to pro-market measures in countries
where there is a competitive wholesale market
25
The theory predicts that quota obligation is more cost-eï¬€ective in promoting renewable technology,
especially those close to commercial stage (such as onshore wind) because it allows competition between
providers. Observations in our sample shows that it is more complex in practice. As shown in Table (2),
17
Figure (2) plots FiT rates against wind capacity installation under standard and vari-
able FiTs. It shows that higher FiT support is not necessarily associated with higher level
of wind development. Figure (3) presents cross-country comparison of the average FiT
rates and average annual wind installation normalized by a countryâ€™s (1) per capita GDP,
(2) total power generation, and (3) estimated wind technical potential. The ï¬?gures show
large diï¬€erences in policy impact in countries with similar remuneration levels. Coun-
tries that provide very high remuneration sometimes achieved very low penetration levels.
Both ï¬?gures (2) and (3) present suggestive evidence of the lack of connection between FiT
remuneration and renewable installation.
5 Empirical Results
5.1 Wind Installation
Table (3) reports various estimators of Equation(8), in which the log of wind installation
is the dependent variable. Columns (1) and (2) are estimators of the OLS and ï¬?xed-eï¬€ects
models. Columns (3) and (4) report the diï¬€erence-GMM results, using the log of wind
capacity (logKit ) and electricity prices (logeit ) lagged 2 to 4 periods in levels, and the log
of FiT prices (logfit ) lagged 1 to 3 periods in levels as instruments; Column (4) introduces
an external instrument (natural gas prices in levels) in addition to the standard GMM
instruments. Columns (5) and (6) correspond to the system GMM results, where we add
the equation in levels with the ï¬?rst-diï¬€erences of logKit and logfit lagged 1 period and
the ï¬?rst-diï¬€erence of logeit lagged 2 periods as instruments (plus natural gas prices as an
exogenous instrument in column (6)). Finally, columns (7) and (8) are OLS and ï¬?xed
eï¬€ects estimators of a static model described in Equation(4) (with no instruments). All
GMM results are robust one-step estimators.
The coeï¬ƒcient of logKi,tâˆ’1 is highly signiï¬?cant in all models and justify the adoption
premium prices for wind in a TGC market does not seem to be lower than FiT subsidies, given the similar
level of wind deployment. One explanation for this is that price uncertainty under quota obligations could
have driven up certiï¬?cate prices to compensate for the risk premium.
18
of a dynamic model. Both OLS and within-group estimators in columns (1) and (2) are
inconsistent. The OLS model ignores the country-ï¬?xed eï¬€ects that are correlated with
both dependent and explanatory variables. The within results fail to account for the
correlation between transformed lagged dependent variable and the transformed error term.
Nonetheless, the OLS and within-group estimators provide upper and lower bounds for a
candidate consistent estimator of the coeï¬ƒcient of the lagged dependent variable (Î±) (Bond,
2002). Note that all GMM results on Î± lie within the boundary of the OLS and within-
group estimators. However, the standard ï¬?rst-diï¬€erenced GMM estimator is found to be
very close to within-group results, suggesting weak instrument and potential ï¬?nite sample
bias (Blundell and Bond, 1998). Results of the diï¬€erence GMM augmented with external
instrument are improved. The system GMM are both higher and better determined than
both diï¬€erence GMM estimators. The overidentifying restrictions are valid in all cases. All
GMM estimators have also decisively passed the Arellano-Bond test for ï¬?rst- and second-
order autocorrelation in the ï¬?rst-diï¬€erence errors implying that the disturbance term (wit )
is ï¬?rst-order serially uncorrelated (with the exception of diï¬€erence GMM with external
instrument in column (4)) and that our moment conditions are valid. The dynamic common
factor restrictions are easily accepted in all but the standard diï¬€erence GMM in column
(3). Taken together, we choose system GMM with the external instrument as the preferred
results.
Several conclusions emerge from the results. First, the coeï¬ƒcient associated with FiT
rates is positive but insigniï¬?cant, implying that higher FiT rates do not necessarily lead to
higher levels of wind installation. This result is stable as we move from one model to another
(sign and signiï¬?cance). There are at least two explanations for the lack of strong connec-
tion between FiT levels and investment. First, as pointed out in IEA(2011), countries with
high remuneration levels may lack the necessary institutional and regulatory environment
to attract investment, and fail to scale-up investment due to these non-economic barriers.
Another plausible explanation is that overly generous subsidies may have driven up invest-
ment costs by allowing (1) ineï¬ƒcient development in less economical sites such as those
with low wind speed, and (2) rent-seeking so that any excessive compensation is captured
19
by higher system costs. Note that the coeï¬ƒcients of FiT rates of the static models (columns
(7) and (8)) are negative. This downward bias supports the second hypothesis that FiT
rates and unobserved investment costs are positively correlated.
Second, the length of FiT contract and guaranteed grid access both have a positive
and statistically signiï¬?cant eï¬€ect on wind capacity growth. On average, a one percent
increase in contract length increases annual wind installation by 0.3 percent. Evaluated
at the mean level of wind annual installation and mean contract length in the sample, a
one standard deviation increase in the length of contract duration increases annual wind
installation by 18 percent, or 20MW annually. The eï¬€ect of grid connection is even more
conspicuous. Providing grid access almost doubles wind installation in one year, ceteris
paribus 26 . Consistent with ï¬?ndings of prior studies (Wiser and Langniss, 2011; Swisher
and Porter, 2006), the result suggests that investors attach high premiums to the certainty
of investment return. A stable market system that encourages longer-term contracting can
often be more cost-eï¬€ective than those in which short-term trade predominate.
Third, higher electricity prices are correlated with lower investment on wind capacity.
However, when we interact electricity price with the dummy variable indicating the ex-
istence of a competitive wholesale market, the coeï¬ƒcient associated with the interaction
term, as shown in Table (4), is positive and statistically signiï¬?cant. The results hold for
all models with the exception of diï¬€erence GMM with external instrument (column (4) in
Table (4)). Based on our preferred system-GMM estimator in column (6) in Table (4), a
one percent increase in electricity prices on average leads to a 0.7 percent increase in wind
deployment in places served with a competitive wholesale market. This result highlights
the importance of electricity market design in renewable development. When FiT rates
are independent from electricity price (as in the case of standard FiT and mostly variable
FiT), the higher the conventional electricity prices, the lower the relative premiums for
renewable electricity, and the less attractive are renewable projects. But in a competitive
26
Note that the dummy variable for grid access is dropped in within-group and ï¬?rst-diï¬€erenced GMM
models because it is time-invariant. But the system GMM is able to measure the impact of grid access as
it includes the levels equation.
20
wholesale market, renewable generators can beneï¬?t from higher conventional electricity
prices by selling directly to the market. As a result, higher electricity prices induce larger
amount of renewable investment.
Fourth, total electricity demand proxied by total power generation has a strong positive
impact on wind incremental investment. A one percent increase in electricity demand is
associated with a 0.7 percent increase in wind installation in the long-run. Coeï¬ƒcients
of the trend variable and most of the year dummies (not reported) are also positive and
statistically signiï¬?cant.
Fifth, to compare the policy impact of standard FiT, variable FiT and RPS policies,
we interact policy dummies with corresponding standard and variable FiT rates, and the
average TGC prices respectively. Table (5) reports the results. Diï¬€erent from FiT rates,
the TGC prices are more likely to be contemporaneously correlated with wind installation.
This is because the price of green certiï¬?cates is determined by the balance between renew-
able target and the supply of renewable energy. When renewable target is binding, higher
wind penetration is expected to result in lower TGC prices. To explore the causal relation-
ship between TCG price and wind investment, we treat the value of green certiï¬?cates as
endogenous and use lagged variables dated 2 to 4 periods as instruments.
As shown in Table (5), the AR(1) coeï¬ƒcient of wind installation of diï¬€erence GMM esti-
mator is no longer signiï¬?cant and is in fact outside of the bounds of OLS and within-group
estimators. The diï¬€erence GMM with external instrument performs better suggesting
that the external instrumental variable does help to better identify the dynamic equation,
however the AR(1) coeï¬ƒcient in column (4) is still only slightly above the within-group
estimator. Arellano-Bond test does not reject the null hypothesis of zero autocorrelation
in the ï¬?rst-diï¬€erenced errors at order one in all speciï¬?cations. The overidentiï¬?cation test
supports the validity of instruments. All models except diï¬€erence GMM (column(3)) pass
the common factor restriction tests. As a whole, we continue to prefer results based on the
system GMM augmented with external instrument.
With the alternative speciï¬?cation, still we do not detect a strong correlation between
FiT remuneration and wind installation - the coeï¬ƒcients associated with FiT rates of either
21
standard or variable FiT policies are positive but insigniï¬?cant. However, the TGC price
does have a statistically signiï¬?cant positive impact on the amount of wind installation. A
one percent increase in a TGC price increases wind investment by 1.7 percent on average.
All other results are highly consistent with previous estimation.
5.2 Wind Generation
We also examine the relationship between FiT rates and wind output based on speciï¬?cation
identiï¬?ed in Equation (6). The econometric results are illustrated in Table (6). Columns (1)
and (3) report the ï¬?xed eï¬€ects estimators27 . Columns (2) and (4) correspond to two-stage
least squares (2SLS) estimation using average natural gas prices as instrument for electricity
prices. In columns (3) and (4), we also introduce the interactions of policy dummies with
FiT rates and TGC prices. Note that despite their being statistically insigniï¬?cant, the
coeï¬ƒcient associated with electricity price is negative in the ï¬?xed eï¬€ects model, and positive
in the 2SLS model. The negative correlation between wind generation and electricity
price could have contributed to the opposite sign on the endogenous variable. The 2SLS
estimators are therefore our preferred results.
Our sixth ï¬?nding is that after controlling for the total cumulative wind capacity, higher
FiT subsidies are not correlated with higher wind generation. In fact, as shown in column(4)
of Table (6), a 1 percent increase in a variable FiT is associated with a 1.9 percent reduction
in wind generation. The result is statistically signiï¬?cant at 10 percent conï¬?dence level. This
ï¬?nding is consistent with the hypothesis that higher ï¬?nancial support induced wind projects
in low quality sites with less favorable wind conditions. Consequently, the wind generation
becomes lower given the same level of wind capacity.
5.3 Robustness Check
First, as pointed out by IEA(2011), it is possible that the growth of renewable is initially
constrained as the supply chain is put in place and administrative systems are streamlined.
Capacity then grows more quickly as the local industry matures and regulators become
27
A Hausman test decisively rejects random eï¬€ects model at any level of signiï¬?cance.
22
more experienced. To measure the learning eï¬€ects, we introduce an additional regressor
that measures the number of years a FiT policy had been put in place in a given country. We
ï¬?nd introducing the measurement of policy implementation horizon has very little impact
on the results.
Second, although labor costs are only a small component of the overall production
costs, it may still aï¬€ect FiT compensation scale and investment decisions. If labor costs
are correlated with both FiT support and wind installation, the elasticity of FiT price will
be underestimated. To explore how important is such an eï¬€ect, we use the average em-
ployee cost in electricity generation sector (NACE 3511) reported in AMADEUS database
maintained by Bureau Van Dijk as an approximation for labor cost of wind generation.
Introducing the additional variable further reduces the sample size, but the basic results
do not change.
The preceding results are also robust to alternative speciï¬?cations assuming FiT re-
muneration as strictly exogenous, or using instruments for wind installation lagged three
periods and earlier to account for possible MA(1) measurement errors.
6 Conclusion
FiTs are rapidly emerging as the most widely used policy instrument to promote renewable
energy deployment in the world. Meanwhile, the cost impacts of FiT policies have become
a primary concern for policy makers, and particularly so in developing countries. Whether
the costs are recovered from ratepayers or taxpayers, rising costs can create both political
and economic pressures. Households in developing countries are particularly vulnerable
to increase in energy prices because energy comprises a higher share of their income than
those in developed countries.
While there exists a large literature on the review of FiT design, empirical studies on the
cost-eï¬€ectiveness of FiT policies are scarce. This paper adds to the literature by estimating
the sensitivity of renewable investment to FiT remuneration levels based on wind data of
35 European countries between 1991 and 2010. Understanding the relation between tariï¬€
23
setting and policy outcome is important not only because it is a critical step to examine
the cost-eï¬€ectiveness of the existing policy design but also it can provide useful lessons for
countries who are contemplating adopting FiT programs.
The performance of a FiT policy can be judged by its impact on a number of pa-
rameters, i.e. installed capacity, energy production, improved energy security, greenhouse
gas reduction and other environmental beneï¬?ts, as well as economic development oppor-
tunities. In this paper, we assess policy performance based on its impact on annual wind
capacity installation and generation. Our analysis shows that although FiTs have worked
to spur investment, high remuneration levels have not necessarily yielded greater levels of
installation. We conjecture that non-economic barriers and the negative correlation be-
tween investment costs and FiT rates could contribute to the weak correlation between
incentives and investment. Countries providing high incentives may happen to be the ones
that have very high levels of non-economic barriers. Excessive subsidies could also provide
rent-seeking opportunities or allow investment in high cost sites.
On the contrary, FiT contract duration and interconnection guarantee are important
determinants of the â€œshadow priceâ€? of FiT support - longer contracts and guaranteed
grid access have induced larger deployment. Because both features are crucial to investor
conï¬?dence in long-term market security, the results suggest that policy certainty is just as
important, if not more than, short-run ï¬?nancial incentives to attract private participation.
Our results also show that the overall electricity market design matters. A competitive
market tends to be more conducive to renewable deployment.
Finally, we show that given cumulative capacity installation, variable FiT rates are
negatively correlated with wind generation. Since wind capacity factor is largely determined
by local wind speed, higher ï¬?nancial incentives could have pushed investment to locations
with less favorable wind conditions. Overall, our analysis provides preliminary evidence of
the dynamic correlation between investment costs and FiT support levels. More in-depth
incidence analysis on who beneï¬?ts from high FiT rates can be an interesting topic for future
research.
24
References
[1] Arellano, M. and S. R. Bond (1991), â€?Some Speciï¬?cation Tests for Panel Data: Monte
Carlo Evidence and an Application to Employment Equations,â€? Review of Economic
Studies 58:277-298.
[2] Arellano, M. and O. Bover (1995), Another Look at the Instrumental-Variable Estima-
tion of Error-Components Models, Journal of Econometrics, 68: 29-52.
[3] Blundell, R. W. and S.R.Bond. 1998. Initial Conditions and Moment Restrictions in
Dynamic Panel Data Models, Journal of Econometrics, 87: 115-143.
[4] Blundell, R.W. and S.R. Bond. 2000. GMM Estimation with Persistent Panel Data:
An Application to Production Functions, Econometric Reviews, 19: 321-340.
[5] Bond. S. R. 2002. Dynamic Panel Data Models: A Guide to Micro Data Methods and
Practice. Center for microdata methods and practice. Working Paper CWP09/02.
[6] Buen. J. 2005 Danish and Norwegian wind industry: The relationship between policy
intruments, innovation and diï¬€usion. Energy Policy 3: 3887-3897.
[7] Cliï¬€ord, E. and M. Clancy (2011). Impact of Wind Generation on Wholesale Electricity
Costs in 2011, SEAI, Eirgrid.
[8] European Wind Energy Association (EWEA). 2009. The Economics of Wind Energy.
[9] Fischer, C. and Preonas, L. 2012. Feed-in tariï¬€s for renewable enregy: eï¬€ectiveness and
social impacts. World Bank Report.
[10] Heiman, M.K. 2006. Expectations for renewable energy under market restructuring:
the U.S. experience. Energy 31:1052-1066.
[11] IEA. 2012. IEA Energy Prices and Taxes Statistics. OECD/IEA, Paris.
[12] IEA. 2011. Deploying Renewables 2011: Best and Future Policy Practice. OECD/IEA,
Paris.
25
[13] IEA. 2008. Developing Renewables: Principles for Eï¬€ective Policies, OECD/IEA,
Paris.
[14] Menz. F. C. and Vachon. S. 2006. The eï¬€ectivneses of diï¬€erent policy regimes for
promoting wind power: Experiences from the states. Energy Policy 34: 1786-1796.
[15] Mulder, A. 2008. Do Economic Instruments Matter? Wind Turbine Investments in
the EU(15). Energy Economics 30(6): 2980-2991.
[16] Palmer, K. 1997. Electricity restructuring: environmental impacts. Forum for Applied
Research and Public Policy 28(6).
[17] Phillips, P.C.B., Sul, D., 2007. Bias in dynamic panel estimation with ï¬?xed eï¬€ects,
incidential trends and cross section dependence. Journal of Econometrics 137:162-188.
[18] Prescott, Edward C. 1971. Investment Under Uncertainty. Econometrica. V(39): 659-
681.
[19] REN21. 2012. Renewables 2012 Global Status Report. Paris: REN21 Secretariat.
Ä±o, P.d. and Gual. M. A. 2007. An Integrated assessment of the feed-in tariï¬€ system
[20] RÂ´
in Spain. Energy Policy 35: 994-1012.
[21] Sawin, J.L. 2006. â€?National Policy Instruments: Policy lessons for the Advancement
and Diï¬€usion of Renewable Energy Technologies Aroudn the Worldâ€?, in AÃŸmann, D.
and U. Laumanns (eds.). Renewable Energy: A Global Review of Technologies, Policies
and Markets. Earthscan, London.
oderholm. P. and Klaassen G. 2007. Wind Power in Europe: A Simultaneous
[22] SÂ¨
Innovation-Diï¬€usion Model. Environmental & Resource Economics 36:163-190.
[23] Zhang, F. 2009. Turkey Renewable Energy Project. World Bank Project Implementa-
tion Completion Report. Washingington DC.
26
Table 1: Summary Statistics
Variable Obs Mean Std. Dev. Within S.D. Min Max AR(1) SYS GMM
Annual wind installation (MW) 740 110.63 369.26 202.09 0 3247 0.736
(0.068)
Annual wind generation (TWh) 653 1.25 4.49 2.51 0 38.55 0.913
(0.024)
FiT rate (euro cents/kWh) 270 6.94 2.62 0.80 1.10 24.11 0.805
(0.086)
TGC price (euro cents/kWh) 29 6.03 2.59 0.86 1.66 11.57
FiT contract length (years) 294 11.52 6.73 4.76 0 25 0.913
(0.028)
Grid access 301 0.91 0.29 0.3 0 1 1
(0.024)
Avg. end-use electricity price (euro cents/kWh) 473 7.72 3.79 1.3 0.89 25.21 0.922
(0.025)
Avg. industrial natural gas price (euro/GJ) 333 5.09 1.83 0.98 12.02 0.651
(0.054)
27
Total electricity output (TWh) 683 100.00 133.28 13.41 1.95 594.69 0.981
(0.010)
Wind power potential (TWh) 575 1611.14 1607.34 0 42.29 5279.59 1
Competitive wholesale market (0/1) 740 0.32 0.47 0.3 0 1 0.946
(0.008)
Table 2: Country Characteristics by Policy Type
Variable Average FiT TGC No Policy Standard FiT Variable FiT Premium FiT
Annual wind installation (MW) 301 298 3.33*** 325 102.60 85.86
(221.81) (326.3) (22.9) (688) (227.68) (125.15)
Annual wind generation (TWh) 2.73 1.93 0.024*** 4.16 0.71 1.89
(6.70) (2.11) (0.18) (8.83) (1.35) (2.19)
FiT rate / TGC price (euro cents/kWh) 6.67 6.03 - 6.69 6.62 4.22
(2.75) (2.59) (3.04) (1.84) (2.37)
FiT contract length (years) 11.52 - - 14.01 9.07 6.88
(6.73) (4.78) (6.24) (9.03)
Grid access 0.91 - - 0.99 1 0.5
(0.29) (0.13) (0) (0.51)
Avg. end-use electricity price (euro cents/kWh) 7.88 9.04 5.74*** 7.79 7.86 8.23
(3.29) (3.86) (4.49) (3.23) (4.07) (1.17)
Avg. industrial natural gas price (euro/GJ) 5.32 5.74 4.5*** 5.61 4.97 4.9
28
(1.74) (2.38) (1.64) (2.02) (1.22) (1.23)
Total electricity output (TWh) 119.65 217.15 48.9*** 152 88.04 70.46
(166.5) (107.2) (67.1) (191) (148.44) (49.93)
Wind power potential (TWh) 1433 3021 1156.4*** 1522 1042 1836
(1553) (1719) (1241.9) (1701) (1062) (1664)
Competitive wholesale market (0/1) 0.51 0.73 0.06*** 0.58 0.24 0.77
(0.50) (0.45) (0.24) (0.49) (0.43) (0.42)
Obs. 301 41 545 165 88 48
Note: Obs. is the number of observations of annual wind installation.â€œNo policyâ€? includes 9 countries that have never implemented wind incentive policies
during the sample period. Ideally, we would only include these countries in the last column. But data on wind potential, electricity and natural gas
prices are not available for most of these countries. - indicates that the numbers are not available. *** indicates the diï¬€erence between â€œNo Policyâ€? and
â€œAverage FiTâ€?is statistically signiï¬?cant at 1% level. For comparison among FiT, *** indicates the diï¬€erence between â€œVariable FiTâ€? or â€œPremium FiTâ€?
and â€œStandard FiTâ€? is statistically signiï¬?cant at 1% level, ** indicates signiï¬?cance at 5%, and * indicates signiï¬?cance at 10%.
Table 3: Determinants of Log of Annual Wind Installation (I)
(1) (2) (3) (4) (5) (6) (7) (8)
OLS Within Group DIF GMM DIF GMM SYS GMM SYS GMM Static Static
+ external IV + external IV OLS Within Group
Lagged wind installation 0.800*** 0.360** 0.380*** 0.488*** 0.583*** 0.635***
(0.0595) (0.133) (0.120) (0.166) (0.071) (0.077)
Log FiT rates 0.0298 0.803 1.396 1.612 0.388 0.620 -0.096 -0.509
(1.370) (1.504) (1.423) (1.372) (1.129) (0.900) (0.604) (2.156)
Lagged log Fit rates -0.312 1.428 1.828* 0.889 0.647 0.200
(1.310) (1.437) (1.101) (1.088) (0.961) (0.918)
Competitive market -0.699 -0.462 -0.492 -0.386 -0.530 -0.589 0.532 0.866
(0.508) (0.358) (0.299) (0.716) (0.386) (0.764) (0.533) (0.588)
lagged Competitive market 0.941** 1.269*** 1.297*** 0.532 1.582*** 0.708
(0.462) (0.419) (0.377) (0.604) (0.515) (0.442)
Log electricity price -1.047 -2.509** -2.858*** -2.564* -0.935 -2.840** 0.878* -4.477**
(0.931) (0.954) (0.836) (1.368) (1.010) (1.327) (0.520) (2.051)
lagged log electricity price 0.558 -2.518 -2.712* 0.533 -0.238 1.911
(0.900) (1.719) (1.415) (1.128) (1.256) (1.212)
29
Log power output 3.051** 3.909** 4.217*** 6.179*** 3.464*** 4.517** 0.941*** -2.184
(1.517) (1.608) (1.391) (1.862) (0.951) (1.803) (0.178) (2.950)
lagged log power output -2.936* -0.276 -0.677 0.115 -3.280*** -4.263**
(1.491) (2.361) (2.244) (1.390) (0.945) (1.819)
Log contract year 1.137 0.733 1.561 2.778* 1.96 2.264** 0.398 -2.188
(1.700) (1.144) (1.552) (1.653) (1.460) (0.946) (0.890) (1.930)
Lagged log contract year -0.473 -0.893 -0.316 -1.911 -1.856 -2.161***
(1.362) (0.903) (0.883) (1.281) (1.309) (0.790)
Guaranteed grid access 1.387*** - - - 1.417 1.801** 1.032 -
(0.270) (1.284) (0.708) (0.657)
Time trend 0.137*** 0.0875 0.112* 0.150* 0.0919** 0.151*** 0.300*** 0.169**
(0.0516) (0.0665) -0.0649 -0.0902 (0.0357) (0.038) (0.070) (0.075)
R-squared 0.863 0.771 0.496 0.565
Obs. 100 100 77 54 100 70 128 128
ï¬?rst-diï¬€erenced errors m1: p-value 0.0098 0.0099 0.0052 0.0198
ï¬?rst-diï¬€erenced errors m2: p-value 0.3999 0.0782 0.8992 0.3790
Sargan test: p-value 0.5018 0.4667 0.8943 0.9912
Common factor restiction: p-value 0.0074 0.4362 0.2179 0.3584
Note: Columns (1) and (2) report results from estimating determinants of annual wind installation identiï¬?ed in equation (8) via OLS and ï¬?xed eï¬€ects model. Columns (3) and (4) report the
diï¬€erence-GMM results, using the log of wind capacity (logKit ) and electricity prices (eit ) lagged 2 to 4 periods in levels, and FiT prices (fit ) lagged 1 to 3 periods in levels as instrument;
Columns (5) and (6) corresponds to the system-GMM results, where equation in levels is instrumented with the ï¬?rst-diï¬€erenced logKit and fit lagged 1 period, ï¬?rst-diï¬€erenced eit lagged 2
periods. Average natural gas prices are used as an exogenous instrument in columns (4) and (6). Columns (7) and (8) are OLS and ï¬?xed eï¬€ects estimators of Equation(4) with no instruments.
Standard error adjusted for clustering on country are reported in parentheses. *** indicates signiï¬?cant at 1% level, **indicates signiï¬?cant at 5% level, *indicates signiï¬?cant at 10% level. Reported
R2 is the adjusted R2 for OLS and ï¬?xed eï¬€ects model.
Table 4: Determinants of Log of Annual Wind Installation (II)
(1) (2) (3) (4) (5) (6) (7) (8)
OLS Within Group DIF GMM DIF GMM SYS GMM SYS GMM Static Static
+ external IV + external IV OLS Within
Lagged wind installation 0.776*** 0.329** 0.326** 0.419*** 0.596*** 0.596***
(0.0590) (0.139) (0.137) (0.158) (0.0716) (0.0713)
Log FiT rates -0.15 0.385 0.581 1.431 -0.116 0.209 0.414 0.00481
(1.218) (1.394) (1.215) (0.970) (1.008) (0.890) (1.607) (0.497)
Lagged log Fit rates -0.294 1.189 1.226 0.129 0.106 -0.115
(1.202) (1.212) (1.031) (1.181) (1.067) (0.922)
Competitive market*Log electricity price -0.254 -0.0729 -0.079 -0.164 -0.166 -0.110 0.312 0.344
(0.244) (0.167) (0.144) (0.233) (0.199) (0.270) (0.251) (0.238)
Competitive market*lagged log electricity price 0.333 0.429** 0.398** 0.0412 0.506*** 0.302**
(0.221) (0.183) (0.179) (0.147) (0.173) (0.139)
Log electricity price -1.02 -2.807** -2.930*** -1.834 -1.267 -2.596*** -4.630** 0.486
(0.856) (0.995) (0.915) (1.201) (1.189) (0.867) (2.191) (0.468)
Lagged log electricity price 0.846 -2.503 -2.515* 1.133 0.876 2.301**
(0.910) (1.663) (1.439) (0.844) (1.218) (0.900)
30
Log power output 2.540* 2.979** 2.983** 5.535*** 2.044 3.173*** -1.731 0.918***
(1.504) (1.358) (1.346) (1.431) (1.444) (1.075) (3.015) (0.153)
Lagged log power output -2.326 -0.375 -0.531 -0.113 -1.808 -2.818***
(1.475) (2.110) (2.148) (1.024) (1.424) (1.060)
Log contract year 1.065 0.495 0.552 2.066** 1.362 2.847*** -1.849 -0.406
(1.633) (1.257) (1.543) (0.992) (1.746) (1.033) (1.556) (0.760)
Lagged log contract year -0.743 -1.198 -1.045 -2.955** -1.487 -2.722**
(1.336) (0.744) (0.779) (1.373) (1.647) (1.047)
Guranteed grid access 1.488*** 2.610* 2.476*** 1.076
(0.269) (1.388) (0.841) (0.685)
Time trend 0.157*** 0.103 0.118 0.199*** 0.146** 0.181*** 0.184** 0.283***
-0.0484 (0.0694) (0.0723) (0.0767) (0.0716) (0.0327) (0.0814) (0.0674)
R-squared 0.85 0.746 0.54 0.498
Obs. 107 107 82 59 107 77 138 138
ï¬?rst-diï¬€erenced errors m1: p-value 0.0089 0.0078 0.0047 0.0103
ï¬?rst-diï¬€erenced errors m2: p-value 0.1764 0.2701 0.4427 0.5019
Sargan test: p-value 0.4867 0.5059 0.7076 0.9766
Common factor restiction: p-value 0.0406 0.7053 0.9801 0.1598
Note: Columns (1) and (2) report results from estimating determinants of annual wind installation identiï¬?ed in equation (8) via OLS and ï¬?xed eï¬€ects model. Columns (3) and (4) report the
diï¬€erence-GMM results, using the log of wind capacity (logKit ) and electricity prices (eit ) lagged 2 to 4 periods in levels, and FiT prices (fit ) lagged 1 to 3 periods in levels as instrument;
Columns (5) and (6) corresponds to the system-GMM results, where equation in levels is instrumented with the ï¬?rst-diï¬€erenced logKit and fit lagged 1 period, ï¬?rst-diï¬€erenced eit lagged 2
periods. Average natural gas prices are used as an exogenous instrument in columns (4) and (6). Columns (7) and (8) are OLS and ï¬?xed eï¬€ects estimators of Equation(4) with no instruments.
Standard error adjusted for clustering on country are reported in parentheses. *** indicates signiï¬?cant at 1% level, **indicates signiï¬?cant at 5% level, *indicates signiï¬?cant at 10% level. Reported
R2 is the adjusted R2 for OLS and ï¬?xed eï¬€ects model.
Table 5: Determinants of Log of Annual Wind Installation (III)
(1) (2) (3) (4) (5) (6) (7) (8)
OLS Within Group DIF GMM DIF GMM SYS GMM SYS GMM Static Static
+ external IV + external IV OLS Within
Lagged wind installation 0.779*** 0.231 0.227 0.381** 0.597*** 0.632***
(0.0596) (0.154) (0.138) (0.188) (0.0622) (0.0675)
Log FiT rates * Standard FiT -0.0546 -0.813 -0.385 0.992 -0.260 0.576 -0.222 -2.518
(1.380) (1.591) (1.526) (1.284) (1.133) (0.959) (0.538) (2.167)
Lagged log Fit rates * Standard FiT -0.416 0.623 0.627 0.0439 0.595 -0.278
(1.326) (1.307) (1.030) (1.415) (1.009) (1.087)
Log FiT rates* Variable FiT 1.908 5.430* 5.639** 2.013 0.991 0.0473 0.258 5.894***
(2.582) (2.964) (2.674) (1.636) (1.940) (1.497) (0.516) (1.368)
Lagged log FiT rates* Variable FiT -2.185 2.035 2.885 2.502** -0.597 0.398
(2.478) (2.650) (2.422) (1.266) (1.759) (1.096)
Log TGC price * RPS -1.507 -2.505 -3.027 -1.765 -5.240** -4.247** -0.234 2.490*
(3.540) (1.817) (1.996) (3.723) (2.242) (2.066) (0.543) (1.377)
Lagged log TGC price* RPS 1.495 2.581 2.531 1.109 6.126** 4.863*
(3.706) (3.275) (3.019) (4.488) (2.778) (2.532)
Competitive market -0.517 -0.118 -0.145 -0.131 -0.355 -0.0318 0.604 0.895
(0.524) (0.361) (0.317) (0.537) (0.446) (0.752) (0.528) (0.571)
Lagged Competitive market 0.898* 1.338*** 1.434*** 0.506 1.189** 0.538
(0.479) (0.424) (0.397) (0.382) (0.457) (0.374)
31
Log electricity price -1.335 -4.031*** -4.466*** -3.016** -1.791** -3.066*** 0.677 -4.823**
(0.843) (1.000) (0.869) (1.267) (0.788) (0.992) (0.509) (1.865)
lagged log electricity price 0.832 -2.851* -3.047** 0.772 1.188 2.496**
(0.811) (1.653) (1.411) (1.083) (0.917) (1.208)
Log power output 3.059* 3.794** 3.608*** 5.637*** 3.281*** 3.295* 0.976*** -1.555
(1.584) (1.599) (1.280) (1.719) (1.046) (1.780) (0.176) (2.752)
lagged log power output -2.923* 0.378 -0.169 0.581 -3.151*** -3.070*
(1.566) (2.155) (1.842) (1.182) (1.006) (1.748)
Log contract year 1.281 -1.041 -0.194 2.286 1.039 2.643** 0.326 -3.794*
(1.875) (1.840) (2.044) (1.676) (1.487) (1.136) (1.004) (1.954)
Lagged log contract year -0.244 -1.08 -0.797 -2.858* -0.43 -2.487**
(1.580) (1.048) (0.818) (1.540) (1.224) (0.975)
Guaranted grid access 1.323*** 1.449 1.578* 0.91
(0.294) (1.377) (0.862) (0.700)
Time trend 0.146*** 0.0587 0.0955* 0.163** 0.141*** 0.147***
(0.0526) (0.0651) (0.0554) (0.0804) (0.0299) (0.0347)
R-squared 0.859 0.776
Obs. 107 107 82 59 107 77 138 138
ï¬?rst-diï¬€erenced errors m1: p-value 0.014 0.0071 0.0053 0.0143
ï¬?rst-diï¬€erenced errors m2: p-value 0.2171 0.2194 0.4317 0.7735
Sargan test: p-value 0.6002 0.5194 0.9751 0.9975
Common factor restriction: p-value 0.0003 0.9606 0.5611 0.7907
Note: Columns (1) and (2) report results from estimating determinants of annual wind installation identiï¬?ed in equation (8) via OLS and ï¬?xed eï¬€ects model. Columns (3) and (4) report the
diï¬€erence-GMM results, using the log of wind capacity (logKit ) and electricity prices (eit ) lagged 2 to 4 periods in levels, and FiT prices (fit ) lagged 1 to 3 periods in levels as instrument;
Columns (5) and (6) corresponds to the system-GMM results, where equation in levels is instrumented with the ï¬?rst-diï¬€erenced logKit and fit lagged 1 period, ï¬?rst-diï¬€erenced eit lagged 2
periods. Average natural gas prices are used as an exogenous instrument in columns (4) and (6). Columns (7) and (8) are OLS and ï¬?xed eï¬€ects estimators of Equation(4) with no instruments.
Standard error adjusted for clustering on country are reported in parentheses. *** indicates signiï¬?cant at 1% level, **indicates signiï¬?cant at 5% level, *indicates signiï¬?cant at 10% level. Reported
R2 is the adjusted R2 for OLS and ï¬?xed eï¬€ects model.
Table 6: Determinants of Log of Wind Generation
(1) (2) (3) (4)
Within Group Within Group Within Group Within Group
+ IV + IV
Log of total wind capacity 0.859*** 0.965*** 0.879*** 0.956***
(0.0464) (0.0897) (0.0381) (0.0664)
Log of FiT rates -0.317 -0.423
(0.303) (0.493)
Log of FiT rates * standard FiT -0.00127 -0.247
(0.41) (0.365)
Log of FiT rates * variable FiT -0.955*** -1.903*
(0.150) (1.134)
Log of TGC price * RPS -0.360* -1.06
(0.181) (0.727)
Competitive market -0.0338 0.191 -0.0569 0.183*
(0.0810) (0.160) (0.0809) (0.0971)
Log of electricity price -0.093 5.396 -0.00541 3.907
(0.171) (6.209) (0.166) (4.051)
Log of contract year -0.262* -0.302 -0.051 -0.139
(0.137) (0.482) (0.196) (0.426)
Time Trend 0.0945*** -0.459 0.0773*** -0.371
(0.0213) (0.443) (0.0153) (0.294)
R2 0.843 0.929 0.805 0.955
Obs. 136 87 145 96
Note: Columns (1) and (3) report the determinants of log of wind generation identiï¬?ed in Equation (6) using ï¬?xed eï¬€ects models.
Columns (2) and (4) correspond to two-stage least squares (2SLS) estimation using average natural gas prices as instrument for
electricity prices. Standard error adjusted for clustering on country are reported in parentheses. *** indicates signiï¬?cant at 1% level,
**indicates signiï¬?cant at 5% level, *indicates signiï¬?cant at 10% level. Reported R2 is the adjusted R2 for ï¬?xed eï¬€ects model.
32
Figure 1 Number of Countries by policy - More countries have adopted FiT
policies
Note: UK in 2010 implemented both FiT and RPS. This graph reports the total number of countries
implementing FiT, RPS and competitive bidding policies in each year. Under a competitive bidding
system, contracts for renewable projects are oï¬€ered to the lowest bid ï¬?rst, followed by the next
cheapest bid, and so on, until the required renewable capacity target is reached.
33
Figure 2 Wind installation by FiT rates
34
Figure 3 Wind Installation v.s. FiT and RPS prices
35