Environmental and energy e�ciency analysis of EU

electricity industry using an heterogeneous Bayesian

dynamic estimator.

Draft

Simona Bigernaa, Carlo Andrea Bollinoa, Maria Chiara D'Erricoa,∗, Paolo Polinoria

aUniversity of Perugia, Economics Department

Abstract

Environmental and energy e�ciency (EEE) is a crucial key in the electricity sector

to make carbon free power generation. The main factors changing the traditionally

largely asset-based industry moving to a new and more complex decentralized gen-

eration system are the technological changes and regulatory interventions.

This paper intends to analyses the relationship of regulatory intervention on the EEE

for the major 18 European Union countries from 2007 to 2014. The novel-ties of the

method is to deal with the cross-country heterogeneity due to unobserved di�erence

across country regulations.

The analysis follows three steps. First, we compute the EEE measures in elec-

tricity sector using the Malmquist Index of Total Factor Productivity through a

∗Corresponding authorEmail addresses: simona.bigerna@unipg.it (Simona Bigerna),

carloandrea.bollino@unipg.it (Carlo Andrea Bollino), mariachiara.derrico@unipg.it(Maria Chiara D'Errico), paolo.polinori@unipg.it (Paolo Polinori)

Preprint submitted to SIE 2018 September 7, 2018

non-parametric approach (Nakano and Managi, 2008). Second, we regress the EEE

indexes derived in the �rst step on the market and environmental stringency ap-

plying a dynamic panel �xed e�ect model. Finally, in the third step, we relax the

homogeneity assumption, assuming the country-variability of regressors' coe�cients

according to an underlying joint distribution. In this framework, the panel �xed ef-

fect estimates of the previous step are used as parameters for the prior distribution.

The obtained Bayes estimator is the best performing (Baltagi et al., 2008; Jobert et

al., 2017).

Keywords: electricity, total factor productivity growth, undesirable output,

market and environmental regulation, Bayes estimator

JEL: codes: C21, L51, L94, O47, Q58

2

1. Introduction

The analysis of the dynamics of Environmental and Energy E�ciency (hereafter

EEE) is attracting growing attention in recent years in both academic and policy

level since innovation and di�usion of more energy-e�cient technologies is a key factor

of the EU 2030 Climate and Energy Strategy to make carbon free power generation5

and increase the energy performance of national system.

EEE in production, transformation and consumption allows to reach European Union

(EU) greenhouse gas reduction target faster (EEA, 2016). EEE is a crucial key in the

transformation sector to make carbon free power generation. Internal and external

factors are changing the traditionally largely asset-based industry moving to a new10

and more complex decentralized generation system. Internal factors refer to techno-

logical changes (Jamasb and Pollitt, 2008) and to the fuel energy mix that deeply

changed in EU countries also due to the widened spread of renewable energy sources

(Krozer, 2013). External factors involve policy and regulatory interventions (Knit-

tel, 2002), changes in consumers' preferences (Stigka et al., 2014) and environmental15

attitude (Bigerna et al., 2016).

This paper intends to contribute to the literature developing a framework to measure

the technical EEE of EU electricity industries taking into account: i) both non-

separable "good" and "bad" outputs; ii) the impact of sector and environmental

regulation; iii) the spatial component in technical e�ciency explanation.20

The methodological approach is based on a two stage strategy; �rst, a non-parametric

3

methods such as Data Envelopment Analysis (DEA) is used to measure the EEE

in electricity sector using the Malmquist Index (MI) of Total Factor Productivity

(TFP), the overall TFP growth is then decomposed into its three components: i)

technological change, ii) pure e�ciency change and iii) scale e�ciency change in25

order to obtain a more complex picture of the e�ects of policy on TFP change. In

the second stage we apply an econometric analysis regressing the measure of e�ciency

derived in the �rst stage on the sector and environmental regulation indicators using

both the frequentist and the bayesian framework.

Three are the novelties of this paper. First the e�ciency valuation of the electricity30

sector considers, along with the electricity production, the greenhouse gas emissions

as the undesirable output (Scheel, 2001; Yang and Pollit, 2009). Second, we enlarge

the boundary of regulation analysis takes into the account, along with the market

regulation indicators, the e�ects on the e�ciency of the environmental policy strin-

gency. Third, we are going to propose a Bayesian analysis of the coe�cient of the35

regulation variables allowing to relax the homogeneity assumption relative to the

cross section regression coe�cients.

The paper is organized as follows. Section 2 presents a literature review, methods

and data used in the analysis explaining �rst the DEA approach used to evaluate the

TFP growth and then the dynamic panel method applied to evaluate the e�ect of40

the stringency of both sector and environmental policy tools and spatial contiguity.

Empirical results are presented and discussed in Section 3. Finally, some concluding

remarks are exposed in Section 4.

4

2. Materials And Methods

2.1. Related literature45

The reform programs in the European Electricity sectors were undertaken for various

reasons: political ideology, improving government �nance, promoting an European

single market, expanding the internal market to network services (Florio, 2014), re-

ducing the operating costs and the retail prices (Joskow, 2008). The reforms have

addressed di�erent aspects of liberalization such as unbundling network from gener-50

ation to retailing, reducing collusion among large companies, eliminating barriers to

entry and transforming individual state owned monopolies into a single competitive

market. The pro-competitive reforms made this traditionally large asset based indus-

try into decentralized generation system. Restructuring, liberalizing and privatizing

the electricity sector in EU countries have been the focus of many studies to eval-55

uate the performance of electricity sector and to assess implementation of reforms.

Numerous papers provide conceptual discussions of electricity market restructuring

and prescribe the appropriate steps to be taken in implementing reforms.

An overview of experiences in several OECD countries where generation segments

has largely been deregulated while transmission and distribution continue to be reg-60

ulated is provided by Al-Sunaidy and Green (2006) and Joskow (2008). Al-Sunaidy

and Green (2006) show that one compelling reason for the reform in the electric-

ity generation is the lack of natural monopoly in this segment which is common

feature of the transmission segment. Joskow (2008) presents the standard liber-

alization prescriptions to successfully reforming the electricity-supply industry and65

5

to promote performance improvements; some of the main steps are privatization

of formerly state-owned and vertically integrated monopolies; vertical separation (or

unbundling) of the sector to prevent cross-subsidization among various segments and

to ensure equal access to the network for all competitors. They also recommend the

horizontal restructuring of the generation segment to allow competition in power70

production, and the integration of transmission facilities with network operation to

create an independent system operator.

Jamasb and Pollitt (2005) discuss the progress of electricity market reform in EU

countries. The authors noted that EU countries reached signi�cant level of compe-

tition through the liberalization process, but the achievement of an European single75

market for electricity if far from being realized.

The link between e�ciency and regulation in the electricity sector has been already

studied by several authors: Steiner (2000), Hattori and Tsutsui (2004), Pollit (2008),

Fiorio and Florio (2013), Pompei (2013), Hyland (2016) and Ajayi et al. (2017).

The ambiguous impact of unbundling vertically-integrated monopolies has been dis-80

cussed by Pollit (2008) which highlights the two potentially opposite e�ects: on one

hand unbundling fosters competition and improves the operational e�ciency, on the

other it implies the loss of scope and coordination economies, increasing operational

costs. The empirical analysis of Hattori and Tsutsui (2004) �nds that competition

and unbundling caused an increase of wholesale prices in OECD countries.85

Fiorio and Florio (2013) investigated the e�ect of reform on a di�erent notion of

e�ciency explained by the level of retail prices in 15 Countries for the years 1978-

2006. Findings are not univocal: unbundling and free entry market regulation are

6

not statistically signi�cant, while privatize ownership is associated to higher electric-

ity prices. Private ownership may decrease ine�ciencies, but this will not necessarily90

translate in lower prices due to the lower elasticity of electricity demand. Authors

highlight the core role of regulation to protect consumers: increasing competition

will result in lower operational costs and bene�ts for consumers only if there is a

stringent regulatory oversight.

To depict the stringency of electricity market regulation and analyze the e�ect on95

the performance of the sector, we restrict the scope to OECD's Product Market Reg-

ulatory (PMR) indicator to obtain an harmonized index at the country level. The

index evaluates at country level how regulation fosters competition reducing state

involvement in business sectors, making it easier for entrepreneurs to create �rms

and to expand them and facilitating the entry of foreign products and �rms.100

Steiner (2000) makes the �rst use of OECD regulatory database to design 8 indica-

tors used to analyse the e�ect of the �rst wave of reforms for the 18 OECD countries.

She �nds that privatization and a lower degree of vertical integration have positive

e�ect on a physical measure of e�ciency derived from the utilisation capacity and

the optimal reserve margin rate. In this vein even Pompei (2013) and Fiorio and105

Florio (2013) used the OECD regulatory indicators as measures of the stringency of

market regulation. Pompei (2013) uses OECD regulatory indicator to look at the

e�ects of the reforms on the TFP growth for the electricity sectors of 18 OECD coun-

tries. He �nds that lower entry barriers promote the shift of technological frontier,

while vertical integration worsen the pure e�ciency, that is the ratio between out-110

put produced and the input deployed. Ajayi et al. (2017) analyze the performance

7

of OECD's power generation sectors in terms of cost e�ciency accounting for the

impact of electricity market structures. The research �ndings suggests there is a sig-

ni�cant impact of OECD's PMR indexes (explaining the market structure) on cost.

Hyland (2016) estimates the impact of restructuring process on wholesale electricity115

prices using as market restructuring variable the index from OECD's Energy Trans-

port and Communication Regulation (ETCR) database. She �nds that restructuring

has, as of yet, no statistically signi�cant impact on electricity prices.

The signi�cant change taking place in European energy markets concurrently with

market restructuring is the increased focus on decarbonization as a policy objective.120

Climate change and greenhouse gas emission have become environmental concerns

more and more pressing in the political agenda, addressing governments to stimulate

eco-innovation. EEE has become one of the core pillar of the EU 2030 Climate and

Energy Strategy. In particular EEE has been identi�ed as a preferential means to

improve the performance of the national energy system as it could help foster sus-125

tainable energy transition and eco-innovation.

International debate has provided several de�nitions of eco-innovation, among them,

the most complete one is given by Kemp and Pearson (2007), according to which

eco-innovation is "the production, assimilation or exploitation of product, product

process, service or management or business that is novel to the organization and130

which results,throughout its life cycle, in a reduction of environmental risk, pollu-

tion an other negative impacts of resources use (including energy use) compared to

relevant alternatives". As a part of 2030 Climate and Energy Package, the EU has

adopted strict greenhouse gas abatement targets and minimum requirements on the

8

portion of electricity generation coming from renewable sources. This has led to135

large-scale deployment of renewable generation technologies, particularly from wind

and solar resources, across EU.

Several studies have investigated the determinants and the role played by di�erent

policy instruments in stimulating and directing technical change in eco-innovation

and more speci�cally in EEE branch. In this current debate Hojnik and Ruzzier140

(2016) and Veugelers (2012) recognized the primary role played by environmental

regulations in spurring eco-innovation, while Bergek et al. (2014) and Johnstone et

al. (2010) identify the variety of the EEE's determinants and highlights the primary

role played by public policies in fostering the rate of introduction of new environmen-

tal technologies to meet sustainable development goals. Moreno et al. (2012) note145

that the increasingly green generation portfolio has ambiguous e�ect on e�ciency

measure in term of retail electricity prices. While the increase in zero marginal cost

power sources in power generation may help to lower prices, the increasing deploy-

ment of intermittent technologies required additional infrastructure investments as

well as embedded subsidies that raise the retailers' electricity prices. Glachant and150

Ruester (2014) highlight how the fragmented national environmental measures as the

support schemes for renewable energies may slow down process toward the European

single market. A �ourishing literature (see among others Costantini et al. (2017)) fo-

cuses on the e�ect of the mix design of the policies and analyses the interactions and

interdependecies between di�erent policy instruments in order to capture the large155

number of linkages in�uencing the dynamic pattern of environmental and technical

e�ciency changes.

9

This paper wants to integrate this two main �eld of investigation, evaluating how the

EEE of electricity sector are a�ected by both market and environmental regulation.

We consider the performance of electricity sector using a pure measure of its technical160

e�ciency to avoid the bias due to the country-speci�c price assessment. Technical

e�ciency is measured through the MI of TFP. The same non-parametric approach

for the electricity sector at country level is used in Nakano and Managi (2008) and

Yang and Pollit (2009).

We shape the MI to incorporate environmental e�ciency, including the greenhouse165

gas emission as bad output. This measure of e�ciency increases as the electricity

produced increases and as the greenhouse gas emissions decrease. This strategy, in a

parametric framework, has been already followed by Ajayi et al. (2017) who investi-

gate the impact of electricity reform on cost e�ciency in the OECD countries using

a multiproduct production function including greenhouse gas emissions as a part of170

the outputs of the electricity generation process.

As the analysis covers di�erent countries and time periods, potential endogeneity

may occur: while prices and investment in the industry may be a�ected by reform,

these variables are likely to a�ect the decision of restructuring. Following the re-

cent literature (Nakano and Managi, 2008; Gugler et al., 2013; Fiorio and Florio,175

2013; Pompei, 2013; Hyland, 2016) we use dynamic panel data techniques to over-

come this shortcoming. Nakano and Managi (2008) measure productivity in Japan's

steam power-generation sector and examine the e�ect of reforms on the productivity

of this industry over the period 1978�2003. In the �rst stage they estimate the Lu-

enberger productivity indicator, in the second step the e�ect of regulatory reforms180

10

is examined using dynamic Generalized Method of Moment (GMM) estimation for

panel data to take into account the serial correlation among e�ciency estimates. The

empirical results show that regulatory reforms in Japan have contributed to produc-

tivity growth in the steam power-generation sector. Gugler et al. (2013) analyze how

market reforms a�ect the investment in the industry of 16 EU countries from 1990 to185

2010. They conclude that regulatory measures a�ecting market (such as introduction

of the wholesale electricity market) foster investment, while measures a�ecting the

incumbent (such as ownership unbundling) decrease investment.

In most of these empirical panel studies with cross country heterogeneity, Fixed

E�ect (FE) models allow the intercept (the unobserved e�ect) to di�er among cross-190

sections in order to account for country speci�c characteristics, but they assume the

complete homogeneity of the slope parameters. However, in most of the empirical

studies the hypothesis of homogeneity of the slope parameters is often rejected in fa-

vor of heterogeneous regressions. According to Maddala et al. (1997), the true values

of the parameters fall in between complete homogeneity and heterogeneity. Param-195

eters are not exactly the same, but there is some similarity between them. Jobert

et al. (2018) study the relationship between air pollution, international trade �ows

and environmental regulation using aggregate data of 55 countries. They relax the

homogeneity assumption of parameters allowing both the intercepts and the slopes

of the linear model to vary across countries using a Bayesian iterative estimator.200

Following Jobert et al. (2018), in a third step we estimate the relationship between

EEE and regulation by means of the Bayesian estimator in order to allow the slopes

coe�cients to vary across countries. To sum up the contribution of this study to the

11

current literature is threefold. We evaluate the EEE and its changes due to both

market and environmental policy stringency. We implement a new measure of e�-205

ciency that takes into account the environmental concerns, including the greenhouse

gas emission as bad output. The endogeneity problem due to managing with time

and country level data is overtaken using dynamic GMM estimation method. More-

over, we use in a third step a Bayesian estimator that allows to account for an high

degree of country heterogeneity in the relationship between e�ciency and regulation210

since we consider country-speci�c regulation coe�cients as random variables drawn

from a common probability distribution.

2.2. Methods

The impact analysis of regulation on the EEE of electricity sector involves two steps.

In the �rst step we compute the TFP based on MI, in the second step we regress215

the computed country-speci�c TFP on the regulation variables in order to evaluate

their impact on the e�ciency scores.

2.2.1. The TFP analysis

The �rst step involves to apply DEA and compute the MI; in the comparative statis-

tics analysis this index allows to evaluate the TFP change of the Decision Making220

Units (DMU) between two time periods using linear program technique. The theo-

retical framework and computation follow the methodology explained in Cooper et

al. (2007, pp:328-338).

12

We deal with a panel dataset (xj;yj)t with j = 1, ..., n refers to the DMUs and t

refers the period. Each observation has m inputs included in the vector xj ∈ Rm

and q outputs denoted by yj ∈ Rq. For each period t the production possibility set

is given by:

(X, Y )t = {(x, y)|x ≥n∑j=1

λjxtj ,0 ≤ y ≤

n∑j=1

λjytj ,

L ≤ eλ ≤ U,λ ≥ 0} (1)

X t = (xt1, ...,xtn) and Y t = (yt1, ...,y

tn) are respectively the input and output matrix

of DMU for period t; λ = (λ1, λ2, ..., λn)T ∈ Rn is the semipositive vector of the225

intensity variables; L and U are the lower and the upper bound for the sum of the

intensities; if (L,U) = (0,∞) technologies have constant return to scale, while if

(L,U) = (1, 1) the model implies variable return to scale.

Each DMU solves four linear programs, in two of them, at time t and t + 1, DMU

minimizes the distance from the e�cient frontier of the same period (the within

score); in the last two, at time t and t + 1, DMU minimizes the distance from the

e�cient frontier referring the time t+ 1 and t respectively (the intertemporal score).

Let be s = {t, t+ 1}, the �rst two within scores are given by the following linear

13

programs:

δs(xj, yj)s = minθ,λθ

s.t.

xsj ≥ Xsλ

(1/θ)ysj ≤ Y sλ

L ≤ eλ ≤ U

λ ≥ 0 (2)

Let be s = {t, t+ 1} and r = {t+ 1, t}, the two intertemporal scores are given by

the two linear programs:

δs(xj, yj)r = minθ,λθ

s.t.

xrj ≥ Xsλ

(1/θ)yrj ≤ Y sλ

L ≤ eλ ≤ U

λ ≥ 0 (3)

θ is the objective value representing the e�ciency radial measure. δ ≤ 1 is the

distance function representing the score, that is the distance between the observation230

and the frontier. When δ = 1 the observation (xj, yj)t belong to the frontier.

14

For each DMU, MI is derived from the e�ciency scores of the four linear programs:

δt((xj, yj)t), δt+1((xj, yj)

t+1), δt((xj, yj)t+1), δt+1((xj, yj)t) and it is the geometric

mean of two e�ciency ratios:

MI =

{δt((xj, yj)

t+1)

δt((xj, yj)t)× δt+1((xj, yj)

t+1)

δt+1((xj, yj)t)

}1/2

(4)

δt((xj ,yj)t+1)

δt((xj ,yj)t)represents the e�ciency change from t to t+1 viewed from the perspective235

of the set of technologies available in t.

δt+1((xj ,yj)t+1)

δt+1((xj ,yj)t)measures instead the change in e�ciency from t to t + 1 assuming as

reference frontier the frontier a time t+ 1.

MI>1 signals progress in the TFP of the DMU from period t to period t + 1, while

MI=1 and MI<1 respectively indicate the non variation and the deterioration of the240

TFP.

MI can be seen even as the product of two factors expressing the catch-up and the

frontier shift term.

MI =δt+1((xj, yj)

t+1)

δt((xj, yj)t)×{

δt((xj, yj)t)

δt+1((xj, yj)t)× δt((xj, yj)

t+1)

δt+1((xj, yj)t+1)

}1/2

(5)

The �rst term is the catch-up e�ect and it expresses the degree at which the DMU

improves or worsens its e�ciency towards the frontier from t to t+1. The term in the245

brackets, the frontier shift term, re�ects the change in the e�cient frontiers between

two periods. The �rst ratio inside the brackets in (5) measures the proportional

change in the e�cient frontier at the data observed at time t, while the second ratio

measures the change in the frontier at the data observed in period t+ 1. Changes in

15

e�ciency may also be associated with the returns to scale. Let be s the period of the250

reference e�cient frontier, r the time referring the observation and δsC((xj, yj)r) and

δsV ((xj, yj)r) the scores obtained under constant return to scale (CRS) and variable

return to scale (VRS) environment respectively. For any combination of s and r it

holds that:

δsC((xj, yj)r) ≤ δsV ((xj, yj)

r) (6)

The scale e�ciency of the observation (xj, yj)r relative to the technology at time s255

can be de�ned by the ratio:

σs(xj, yj)r =

δsC((xj, yj)r)

δsV ((xj, yj)r)(7)

σs(xj, yj)r is always lower or equal than one; if σs(xj, yj)r = 1, (xj, yj)

r is posi-

tioned on the most productive scale size region of the technology (X, Y )r. σs(xj, yj)r

is an e�ciency indicator showing how far (xj, yj)r deviates from the point of tech-

nically optimal scale of operation.260

Given the scale e�ciency ratio, the MI between two period t and t + 1 in the CRS

environment (MIC) can be rewritten as the product of three component using MI in

the VRS environment (MIV )1:

MIC = MIV ×[σt(xj, yj)

t+1

σt(xj, yj)t× σt+1(xj, yj)

t+1

σt+1(xj, yj)t

]1/2(8)

The term in the squared brackets in (8) represents is the scale e�ciency term ex-

1See the Appendix

16

pressed as the geometric mean of the two scale e�ciency ratios, one relative to the265

technology at time t and the other relative to the technology at t+1 (see Ray (2004)).

We used the indirect approach (Scheel, 2001) to derived an e�ciency measure consis-

tent with the presence of bad output, that is increasing as the electricity production

increases and when the greenhouse gas emissions decrease. The indirect approach is

based on transforming the values of the undesirable output (Y u) using a monotone

decreasing function f such that the transformed data can be included as normal de-

sirable outputs. The transformation we used is f(Y u) = −Y u and refers the additive

inverse model (ADD)2. Transformation generates the same production possibility set

derived from the input approach where the undesirable outputs are incorporating as

2We apply also the multiplicative inverse model where the function transforming the bad unde-sirable output in the good one is f(Y u) = 1

Y u . Both models generate the same rank of DMU interm of e�ciency. The DEA results referring the inverse multiplicative model are available underrequest.

17

inputs3:

(X ′, Y d)t = {(x′, yd)|x′ ≥n∑j=1

λjx′tj ,0 ≤ yd ≤

n∑j=1

λjydtj

L ≤ eλ ≤ U,λ ≥ 0} (9)

where:

j = 1, ..., n are the DMU

x′j ∈ Rm+1 is the input vector and x′

j = (f(yuj ),xj)

f(yuj ) is the bad output270

xj ∈ Rm is the original input

ydj ∈ Rq−1 is the vector of desirable output.

3An alternative method is the direct approach (Färe et al., 1989; Chung et al., 1997) wherethe linear optimization relaxes the strong disposability assumption: since it is impossible to reduceundesirable output without reducing the desirable one, the output vector are weakly disposable,while only the subvector of the desirable output is strongly disposable. The resulting technologyset is estimated by:

(X,Y u, Y d)t = {(x, yu, yd)|x ≥n∑

j=1

λjxtj ,y

u =

n∑j=1

λjyutj ,0 ≤ yd ≤

n∑j=1

λjydtj ,

L ≤ eλ ≤ U,λ ≥ 0}

where:j = 1, ..., n are the DMUxj ∈ Rm is the input vectorf(yu

j ) is the transformed bad output

ydj ∈ Rq−1 is the vector of desirable output.

The undesirable outputs productivity counterpart to the MI was proposed by is the Malmquist-Luenberger (ML) productivity index.In order to make possible comparisons and evaluate the robustness of the three approaches, resultspertaining the ML index are available under request.

18

2.2.2. The regulation e�ect analysis

The second step involves the analysis of the relationship between e�ciency measures275

derived in the �rst step and the industrial and environmental regulations using GMM.

We apply four di�erent linear models, in the �rst model, as we want to study the

impact of regulation on the overall e�ciency measure, we use as dependent variable

the TFP change given by MI; in the remaining three model we want to analyze

the speci�c e�ect of regulation on the three components of MI (the scale, the pure280

e�ciency and the technological change).

There are well-known problems in using the TFP index as a dependent variable

in econometric speci�cations. Although this index does not su�er from boundary

problems, such as those for DEA e�ciency scores, its estimates are seriously a�ected

by serial correlation. As suggested by Simar and Wilson (1999), we control for serial285

correlation and correct the bias of estimates implementing a bootstrap procedure

in the �rst stage. Moreover, the dynamic GMM model allows us to eliminate the

serial correlation. We set up a panel analysis and we use Arellano-Bond estimation

applying the GMM in the �rst di�erences (Arellano and Bond, 1991). The traditional

assumption in data generating process are as follows (Roodman, 2009):290

� We account for country's speci�c �xed e�ect. This argues against cross-section

regressions, which must essentially assume �xed e�ects away, while it argues

in favor of panel analysis, where variation over time can be used to identify

parameters.

� The process is dynamic since the variable on the left hand side (the MI) is295

19

regressed on its own past realizations which are correlated with the �xed e�ect.

� There may be heteroskedasticity, that is the idiosyncratic disturbances (the

errors purged by the �xed e�ects) may have country speci�c patterns of het-

eroskedasticity but are uncorrelated across countries.

� The only available instruments are internal, they are drawn from the within300

dataset and they are the lags of the instrumented (dependent) variables.

Given these assumptions, the model is the follow:

TFPi,t = αTFPi,t−1 + βxi,t + εi,t

εi,t = µi + υi,t

E[µi] = E[υi,t] = E[µi, vi,t] = 0 (10)

where:

i = 1. . . 18 countries.

t = 2006...2014.

TFPi,t id the total factor productivity index305

TFPi,t−1 is the lagged dependent variable.

xi,t is the set of regressors including the regulation variables and the set of controlling

variables

εi,t is the error terms including:

vi,t the idiosyncratic error and310

µi the �xed e�ect.

20

This model may incur in endogeneity problems due to the presence of lagged depen-

dent variable TFPi,t−1 as regressor correlated with the �xed e�ect error term µi. We

transformed the model in order to remove the �xed e�ect as follows:

∆TFPi,t = α∆TFPi,t−1 + β∆xi,t + ∆vi,t (11)

We start the analysis investigating the stationary properties of time series and their315

possible cointegration. If cointegration occurred, even if all the variables in �rst

di�erences are stationary, the �rst di�erence regression model above would be mis-

speci�ed without the error-correction term. We perform two unit root tests for both

the dependent variable and all the explanatory variables. First, we apply the Harry-

Tsavalis test to verify the null hypothesis that each series share a unit root common320

for all panels. We always reject the null hypothesis of the presence of unit root for

all series except for the market regulation variable in level. Cointegration requires

the equation in level to be balanced. As pointed out by Granger (1981), if there is

cointegration, it is not possible for the dependent variable, the TFP, to be covariance

stationary and for the explanatory variable, the regulation variable, to be a random325

walk, because their linear combination can not be a stationary process. Therefore,

the unbalanced equation already gets rid o� the cointegration hypothesis.4

4We perform cointegration test applying the Westerlund error-correction-based panel cointegra-tion tests, just to make sure. Given only 8 periods, test could be applied only for the linear modelwhere the regulation explanatory variables were included without any lags. The null hypothesis ofno cointegration is not rejected against the alternative that the panel is cointegrated as a whole.

21

First di�erence transformation purges the �xed e�ect term but it does not erase the

dynamic bias and the correlation between ∆TFPi,t−1 and ∆vi,t still remains through

vi,t−1. As suggested by Arellano and Bond (1991), if the idiosyncratic errors vi,t are330

not serially correlated, we can use as instruments for ∆TFPi,t−1 its past realizations

in level starting from TFPt−2 since TFPi,t−2 is correlated with vi,t−2 but not with

vi,t and vt−1.

We run �rst di�erence regression using as dependent variable the TFP indexes de-

rived in the ADD model. Moreover, we apply the same regression for the three335

components of the MI: the scale e�ciency, the pure e�ciency and the technolog-

ical change. Possible heteroskedasticity problems are overcome using the feasible

GMM estimator making the estimates for the standard errors more robust (Rood-

man, 2009).

2.2.3. The Bayesian analysis340

In the FE model, the intercept (the unobserved e�ect) is allowed to di�er among

cross sections in order to account for country speci�c characteristics, but it assumes

a complete homogeneity of the slopes of the coe�cients. In this respect, Hsiao (2003)

shows that in case where the true model's intercept and the slopes are all hetero-

geneous, an estimation using individual intercepts with common slopes may lead to345

wrong conclusions. In order to bypass the assumption of common slopes, models

usually consider each country separately and estimating for each time series the dis-

tinct slopes parameters. According to Maddala (1997), the true values of parameters

fall somewhere in between complete homogeneity and complete heterogeneity. In the

22

relationship between EEE and regulation there may be some degree of country het-350

erogeneity arising from di�erent factors: there is some similarity between parameters

but they are not exactly the same; to not consider this heterogeneity may become

an important missing part of the model equation. One way of allowing this degree

of similarity is to use a Bayesian framework and assume that the slope parameters

are random variables, all come from a joint probability distribution with a common355

mean and non-zero covariance matrix, the Bayesian estimates are then given by the

weighted average of the separate time series estimates for each cross-section and

the pooled panel data estimate. More speci�cally, separate estimation are shrunk

toward the overall pooled estimation, hence the name Shrinkage estimators. Accord-

ing to Maddala (1997) and Baltagi et al. (2003) the Bayes shrinkage estimator best360

performs comparing with homogeneous panel estimators. Furthermore, another ad-

vantage of the Bayesian framework is that, although can di�er across countries, the

number of estimates is reduced, thanks to the random speci�cation of coe�cients. In

this framework, the Arellano-Bond estimates of previous step are used as the mean

and the variance of the prior probability distribution of coe�cients:365

� Prior distribution

β ∼ Normal( ˆβAB, V ar( ˆβAB))

23

� Likelihood

y ∼ Normal(β, σ2)

� Posterior distribution

β|yi ∼ Normal

(ˆβABσ2 + ynV ar( ˆβAB

σ2 + nV ar( ˆβAB

)

where ˆβAB and V ar( ˆβAB) are the Arellano-Bond parameter estimates derived in the

previous step.

2.3. Data

We have a balanced panel dataset of 162 observations, 18 countries and 9 years.

The �rst step computation runs using as input the fuels, the installed capacity and370

the labor, while the output vector concerns the greenhouse gas emissions and the

electricity produced. Data refers to 18 EU countries between the 2006 and the 2014.

The annual data representing the fuels, the raw materials used in electricity gener-

ation, are sourced from Eurostat - Energy Database � (Supply, transformation and

consumption of electricity - Nrg100a) and refer to the "Transformation input used by375

conventional thermal power stations". From the same database we also collect data

referring the "Installed capacity" (Infrastructure - electricity - Nrg113a) thought as

24

a proxy for capital. The labor factor is instead proxied by the number of employees

of electricity sector using the International Labor Organization (ILO) Database and

the Amadeus Dataset jointly. Since ILO Database o�ers the aggregated number of380

"Employees working in the mining and quarrying, electricity, gas and water supply

sector" (activity B-D-E in the ISIC REV.4 classi�cation) we derived an estimate of

the labor force employed in the electricity sector using the share computed through

the Amadeus Dataset.

The variable representing electricity output is the "Total net production" and, again,385

it is sourced from the Eurostat Energy Database (Complete energy balances - Nrg105a).

Data concerning green house gas emissions (GGEs) are collected from OECDDatabase.

This latest variable is the undesirable output from electricity generation which need

to be incorporated into the production model and then minimized. Table 1 reports

the main descriptive statistics for the variables used in the e�ciency analysis.390

[Table 1 here.]

The key explanatory variables for the second step are the overall regulatory index

for the electricity sector and the environmental policy stringency. Both come from

the OECD database.

The overall regulatory index (Regulation) is country speci�c and measures the rate395

of regulation in the electricity sector; given the multidimensionality of regulation, it

is given by the aggregation of four components:

� Entry regulation: the presence of barriers to entry such as the presence of lib-

eralized wholesale market for electricity, the access conditions to the electricity

25

transmission grid;400

� Public Ownership: the percentage of shares owned, either directly or indirectly,

by the government in the largest �rm in the electricity sector;

� Vertical integration: the degree of vertical separation between a certain segment

of the electricity sector and other segments of the industry;

� Market structure: the market share of the largest company in the electricity405

industry.

The Environmental Policy Stringency index (EPS) is a country-speci�c measure

de�ning the degree to which environmental policies put an explicit or implicit price

on polluting or environmentally harmful behavior of the whole economy. The index

ranges from 0 (not stringent) to 6 (highest degree of stringency).410

As for the sector regulation, even for the EPS index the key aspect is the the multidi-

mensionality due to the multitude of instruments that can be used to design environ-

mental policy; following the taxonomy developed by De Serres et al. (OECD, 2010),

the index is based on the degree of stringency of 14 environmental policy instruments

primarily related to climate and air pollution (market based and not). The overall415

EPS index can be decomposed in two elements: the market-based and non market-

based indexes. The EPS market based component refers to the e�ectiveness of price-

based support policies such as Trading scheme (i.e. Emissions Trading Scheme for

CO2, Renewable Energy Certi�cates), subsidy for environmentally-friendly activities

(i.e. Feed-In Tari�s), taxes and charges directly applied to the pollution source (i.e.420

26

tax on emissions of NOx), taxes and charges applied on input or output of a produc-

tion process (i.e. Diesel tax), Deposit-refund systems (i.e. Deposit Refund Scheme

for beverages). EPS non-market based refers instead to quantity-based policies such

as command and control regulations (i.e. Emission Limit Value for NOx for large

size coal-�red plants) and technology and support policies (i.e. Government Research425

and Development expenditures).

We include a set of control variables, apart from the time dummies, constituted by

the principal component of fuelmix, the intensity of Research and development ac-

tivities (R&D intensity) and Combined Heat and Power (CHP) penetration rate.

The principal component of fuelmix (pcaFUELMIX) comes from the principal com-430

ponent analysis of the shares of sources used in each countries in the electricity

generation process. Those shares were collected from the Eurostat Database. Ac-

cording to Canes (2012) and Petterson (2012) the pcaFUELMIX can be thought as

a proxy of both the technology changes employed at country-level and the relative

price changes of the fuels used in the generation process.435

The R&D intensity comes from International Energy Agency database and it is

used as an aggregated measure of all the factors proxy the unincorporated technical

progress.

The CHP penetration rate (CHP penetration) comes from International Energy

Agency database (2010) and measures the amount of electricity produced by CHP440

plants which simultaneously generate electricity and useful heating and cooling from

the combustion of a fuel or a solar heat collector. Since countries in our sample show

di�erent rate of penetration of this technology, we use the CHP as control variable.

27

Table 2 reports the complete set of descriptive statistics for all the variables uses in

the econometric analysis.445

[Table 2 here]

3. Results and discussion

Tables 3 and 4 show the results of the �rst step analysis that are the estimates of MI

and their single components (the frontier shift, the scale e�ciency and pure e�ciency

change) referring the conventional power station. In order to reduce serial correlation450

across the estimated index, we perform bootstrap procedure in the linear programs

setting 200 replications and α equal to 0.05.

[Table 3-4 here.]

Belgium proves to be the country with the best performance, followed by France.

Indeed, this result is not surprising, as these two countries have heavily restructured455

their generation segment plants over the period under investigation as witnessed by

the frontier shift and the scale components that explain the largest part of their TFP

growth.

Before discussing estimation results, it is useful to focus on the �gure 1 which presents

the scatter plots of level of stringency of both sector and environmental regulation460

(Regulation and EPS variables) at the beginning and end of the sample period.

[Figure 1 here.]

28

The lines dividing the graphs' area represent the non-weighted averages for the two

references years and distinguish four di�erent groups of countries. The �rst sub-�gure

is the sector regulation scatter plot whose average lines lie at 2.76 for the 2006 and465

1.89 for the 2013, meaning that the average market regulation rate decreases in the

European regions. The South-West quadrant represents the countries whose sector

regulations have always been lower than the averages in both periods and it includes

Austria, Finland, Hungary, United Kingdom, Germany, Portugal and Spain. In the

opposite quadrant there is the group of EU members with persistently high sector470

regulation, it encompasses France, Greece, Denmark, Ireland, Poland, and Slovakia.

Unless the unbundling and liberalization, these countries have yet been character-

ized by a persistently public ownership of generators company. In the remaining two

quadrants there are the countries whose sector regulation change their stringency

during the overall period. Italy and Belgium undertook a meaningful deregulation475

process (the South-East quadrant) since their indexes are signi�cantly lower than the

average recorded in 2013. In Netherlands, Czech Republic and Sweden, the countries

lying in the North-West quadrant, the deregulation changes have been smaller, even

if Netherlands is quite close to the average of 2013.

The second sub �gure represents the scatter graph for the stringency of environmental480

regulation. The South-West quadrant of persistently low environmental regulation

encompasses Ireland, Greece, Belgium, Poland and Hungary whose EPS indexes are

signi�cant lower than the averages for the two reference periods (2.66 for the 2006

and 2.89 for the 2013). The opposite North-East quadrant depicts a large group

of EU members which during the overall period have always implemented policies485

29

fostering a sustainable energy transition. Starting from Denmark, with the most

stringent environmental regulation, we mention Netherlands, France, Finland, Italy,

Germany, Sweden and Austria. The North-West quadrant includes the countries

with an increasing EPS indexes: Slovakia and United Kingdom while the last group

of countries with decreasing EPS index encompasses Czech Republic, Spain and Por-490

tugal.

We analyze the trends of estimated TFP among the eight groups of countries dis-

cussed above (Figure 2a-3d). Both for the sector and the environmental regulation

the e�ciency scores within the same groups of countries do not show a unidirectional

pattern across the years. Moreover, for each group we can not identify a countries495

whose performance has always been higher than the others. Going forward the years,

MIs intersect many times, showing how each country follows a speci�c dynamic re-

gardless the common stringency of regulation. Given these results, in the second step

we want to investigate if the assumed relationship between e�ciency and regulation

is signi�cant.500

We perform regression using Arellano-Bond method. Results are presented in Tables

5-8.

We �rst apply a dynamic panel data model in the �rst di�erences, industrial and

environmental regulation indicators are used as explanatory variables of the TFP.

Since the lagged �rst di�erence in the right hand side, ∆TFPt−1, is correlated to the505

�rst di�erence of idiosyncratic error ∆vt trough the term vt−1, we use as instruments

all the past realizations in levels starting from time t − 2. If there is not serial

correlation across vt, the vectors TFPt−2, TFPt−3,... and TFP1 are good instruments

30

for ∆TFPt−1 since they are not correlated neither with vt nor with vt−1.

Same methodology is then applied to the sub-components of e�ciency measures: the510

Frontier shift (Tech), the scale e�ciency change (Scale) and pure e�ciency change

(Pe�).

[Tables 5-8 here.]

Before discussing the results of this analysis, it is useful to note the importance of the

lagged dependent variable (L.TFP), which is always signi�cant and negative, show-515

ing a convergence process in which larger variations in the lagged period favor lower

variations in TFP in the next period. This justi�es our choice of dynamic panel data

model. Using the �rst di�erence GMM estimator and the lagged dependent variable

obliged us to discard two time periods; thus, our observations were reduced from 126

to 108.520

As we assume only one endogeneous regressor, to check the power of instruments

Bound et al. (1995) recommend to look at the F-statistic and the partial R2 in the

�rst regression. In all the models we reject the null hypothesis that instruments

jointly do not signi�cantly �t better than the restrict model. Moreover, the F-

statistics in all models are greater than 10, erasing any concern about the weakness525

of instrument used in the �rst stage regression (Staigeret and Stock, 1997).

To investigate the second requirement imposed to the instruments, their orthogonal-

ity to the error process, �rst we look at the p-value of the Hansen test for the validity

of the moment conditions. In all models we can not reject the null hypothesis of the

joint validity of the orthogonality conditions. However, controlling for endogene-530

31

ity may be o�set by the large number of instruments that may over-�t endogenous

variables and fail to expunge their endogenous components, weakening the power

of the Hansen test. Arellano�Bond test for autocorrelation in the residuals in �rst

di�erences may have greater power than the Hansen test to detect if the lagged instru-

ments are invalid due to autocorrelation. If the vi,t are themselves serially correlated,535

TFPi,t−2 is correlated with ∆vi,t = vi,t − vi,t−1 through the term vi,t−1 (Roodman,

2009). Thus we check the Arellano-Bond test for the second order autocorrelation of

∆vt. If ∆vt and ∆vt−2 are not correlated, then vt−2 is not correlated neither with vt

nor with vt−1 and TFPi,t−2 and the longer lags are valid instruments for ∆TFPi,t−1.

Our set of instruments to lags 2 and longer is justi�ed by the Arellano-Bond test540

since in all the models we can not reject the null hypothesis of no second order serial

correlation. We include in the model time dummies in order to control the year

e�ects (time-series trends) and to prevent the contemporaneous correlation. Auto-

correlation test and the robust estimates of the coe�cient standard errors assume no

correlation across countries in the idiosyncratic disturbances; time dummies make545

this assumption more likely to hold removing the universal time-related shocks from

the errors.

Table 5 shows the impact of both overall regulation of the electricity sector (Reg-

ulation) and environmental policy stringency (EPS) with its sub-indicators (Tax,

Trading scheme, Feed in tari�s, Standard and R&D subsidies) on TFP growth. Reg-550

ulation is always signi�cant and negative; more precisely, we �nd that one point

increase in the OECD sector regulation index produces a reduction of TFP between

2.5% and 3.8%.

32

Our results con�rm the �ndings of previous literature (see among others Pompei

(2013); Teece et al. (1994) and Newbery (1997)) in which the stringency of regula-555

tions is negatively correlated to the growth of TFP. Deregulation and most likely

unbundling favor the corporate coherence and the specialization of companies and

help establish an environment of increased competition among fuel input suppliers.

This condition could foster better quality provision of fuels and indirectly improve

the e�ciency of power plants.560

The EPS index has instead positive and signi�cant e�ects on the overall performance.

Given the growing interest in understanding the role played by the di�erent com-

binations of the available environmental policy instruments in directing e�ciency

change, from the model 2 to model 6 we decompose the EPS variable in its main

components in order to evaluate their possible e�ects of the mix design and instru-565

ment interactions. All the di�erent instruments preserve a positive e�ect on TFP

except for Standards whose coe�cients are not signi�cant.

Tables 6, 7 and 8 report the results for the three models where regression pertains

each component of the TFP. On the whole, we observe the absence of a unidirectional

impact of market and environmental regulation when we decompose the dependent570

variable in its three components. This highlights the increasing complexity of the

policy interventions for eco-innovation and how the �nal outcomes of regulation are

becoming more and more di�cult to predict. The coe�cient of sector regulation

persists to be negative for the catch-up e�ect and the scale terms but is not more

signi�cant for the shift in the e�cient frontier. This means that public ownership,575

vertical integration and entry barriers are instruments that push production to bet-

33

ter use inputs and to reach the point of technically optimal scale of operation.

EPS coe�cient keeps to be positive for the scale and the pure e�ciency terms, but

becomes negative for the technology innovation term. Moreover, when we analyze the

impact of di�erent policy tools on the technology innovation, we �nd that the mixed580

instruments a�ect e�ciency in opposite directions. The price-based policies, such as

feed-in tari�s and tax, negatively a�ects eco-innovation, the non-market based tools

seems instead to be e�ective in spurring e�ciency. For all the three components

the coe�cient of the CHP penetration variable is signi�cant and consistent with the

�nding of Yang and Pollit (2009) analysis where an increase in CHP facilities requires585

more fuel for heat and thus a control for a possible downward bias in e�ciency is

needed. For this reason the coe�cient is negative and signi�cant.

Tables 9-12 show the results of Bayesian regression where the slope parameters are

allowed to vary across countries. In this framework we are able to derive the en-

tire posterior distribution of coe�cients. The signs and the values of the Bayesian590

estimate con�rm the results derived in the Arellano-Bond step. The mean of the

posterior distribution of the market regulation coe�cient is negative for the TFP

and its three main sub-components, while the mean of posterior distribution of EPS

is always positive. Figures 4a-5d provide graphical summaries and convergence diag-

nostics for simulated posterior distributions (MCMC samples) of model parameters595

after Bayesian estimation. For each group of countries having di�erent evolution of

regulation we chose the Kernel density of a single nation. The �rst graph plots the

Kernel density distributions of market regulation coe�cients for the TFP of Czech

Republic, Greece, Belgium and United Kingdom; all probability distributions are

34

centered at a negative value close to -0.031, that is the posterior mean of the joint600

normal distribution showed in table 9. The second �gure plots instead the Kernel

density distributions of environmental regulation coe�cients for the TFP change of

United Kingdom, Poland, Portugal and Greece. In this case the Kernel densities are

centered at values close to 0.05 that is the posterior mean of the joint distribution

of EPS coe�cients.605

4. Conclusions

In this study we investigate the relationship between the stringency of market and

environmental regulation and the TFP growth of the electricity sector in 18 EU

countries. Along with the traditional technical productivity, the analysis of e�ciency

takes into account the environmental sustainability of electricity sector developing an610

index able to measure environmental performance though as the ability of reducing

greenhouse gas emissions. We also tested the in�uence of market and environmental

regulation and spatial contiguity on the overall TFP and its three main components.

We found that market and environmental regulation have not univocal impacts

among the aggregated measure of e�ciency and its components. The stringency615

of sector regulation has signi�cant and negative impact on the overall TFP the pure

e�ciency and the scale e�ciency index, but is not signi�cant for the shift of techno-

logical frontier. EPS index, along with its sub-indicators, negatively a�ects the shift

of the e�cient frontier, but has signi�cant and positive e�ect on the scale e�ciency

term. Our analysis tried to develop an appropriate framework in order to capture the620

35

interplay of the wide range of environmental policy instruments and highlighted how

the prices-based tools (represented by the market-based EPS) may not be supportive

drivers in foster EEE. In particular, only the scale e�ciency is positively encouraged

by stringency of market-based environmental policy instruments, on the other hand

the same policy tools counteract the technology innovation change. Results disclose625

the increasing complexity of the policy interventions to trigger sustainable technical

e�ciency and the latent tensions among the di�erent dimensions of the institutional

settings. The Bayesian framework was �nally applied in order to account for a cer-

tain degree of country heterogeneity among the EEE-regulation relationship. The

last analysis was robust with the previous �ndings and allowed to recover the di�erent630

posterior distributions of the slope parameters across countries.

References

Agha, S. B and V. Lionel (2009). �Estimation strategies for spatial dynamic panel

using GMM. A new approach to the convergence issue of European regions.� Spatial

Economic Analysis, 5(2):205�227635

Ajayi, V., T. Weyman-Jones and A. Glass (2017). �Cost e�ciency and electricity

market structure: a case study of OECD countries.� Energy Economics, 65:283�

291.

Al-Sunaidy, A. and R. Green (2006). �Electricity deregulation in OECD (Organiza-

tion for Economic Cooperation and Development) countries.� Energy,31(6):769�640

787.

36

Arellano M. and S. Bond (1991). �Some Tests of Speci�cation for Panel Data: Monte

Carlo Evidence and an Application to Employment Equations.� The Review of

Economic Studies, 58(2): 277�297.

Arellano, Manuel. �Panel data econometrics.�, 2003, Oxford University Press, 1st645

ed., New York, .

Baltagi, B. H., G. Bresson, J. M. Gri�n and A. Pirotte (2003). �Homogeneous, het-

erogeneous or shrinkage estimators? Some empirical evidence from French regional

gasoline consumption.� Empirical Economics 28: 795�811.

Bergek, A., C. Berggren and KITE Research Group (2014). �The impact of envi-650

ronmental policy instruments on innovation: A review of energy and automotive

industry studies.� Ecological Economics, 106:112�123.

Bigerna, S., C. A. Bollino and S. Micheli (2016). �Smart Grids and Consumer At-

titude Toward Sustainable Development.� Journal of Promotion Management,

22(4):573�587.655

Bound, J., D. A. Jaeger and R. M. Baker (1995). �Problems with instrumental vari-

ables estimation when the correlation between the instruments and the endogenous

explanatory variable is weak.� Journal of the American Statistical Association,

90(430):443�450.

Chung, Y. H., R. Färe and S. Grosskopf (1997). �Productivity and undesirable660

outputs: a directional distance function approach.� Journal of Environmental

Management, 51(3):229�240.

37

Cooper, W. W. and Seiford, L. M. and Kaoru T., �Data envelopment analysis.

A Comprehensive Text with Models, Applications, References and DEA-Solver

Software.� Springer, New York, 2007, 2nd edition.665

Costantini, V., F. Crespi and A. Palma (2017). �Characterizing the policy mix and

its impact on eco-innovation: A patent analysis of energy-e�cient technologies.�

Research Policy, 46(4):799�819.

Färe, R., S. Grosskopf and C. A. K. Lovell and C. Pasurka (1989). �Multilateral

Productivity Comparisons When Some Outputs are Undesirable: a Nonparametric670

Approach.� The Review of Economics and Statistics, 71:90�98.

Fiorio, C. V. and M. Florio (2011). �Would you say that the price you pay for

electricity is fair? Consumers' satisfaction and utility reforms in the EU15.� Energy

Economics, 33(2):178�187.

Fiorio, C. V and M. Florio (2013). �Electricity prices and public ownership: Evidence675

from the EU15 over thirty years.� Energy Economics, 39:222�232.

Florio M. (2014). �Network Industries and Social Welfare.� Economics of

Energy and Environmental Policy, 3(2):161�165. ISSN 21605882, 21605890.

http://www.jstor.org/stable/26189283.

Ghisetti, C.,A. Marzucchi and S. Montresor (2015). �The open eco-innovation680

mode. An empirical investigation of eleven European countries.� Research Pol-

icy, 44(5):1080�1093.

38

Glachant, J. M. and S. Ruester (2014). �The EU internal electricity market: Done

forever?� Utilities Policy, 31:221�228.

Granger, C. W. J. (1981). �Some properties of time series data and their use in685

econometric model speci�cation.� Journal of econometrics, 16(1):121�130.

Gugler, K., M. Rammerstorfer and S. Schmitt (2013). �Ownership unbundling and

investment in electricity markets. A cross country study.� Energy Economics,

40:702�713.

Hsiao, C. (2003) �Analysis of Panel Data.� Cambridge University Press.690

Hattori, T. and M. Tsutsui (2004). �Economic impact of regulatory reforms in the

electricity supply industry: a panel data analysis for OECD countries.� Energy

Policy, 32(6):823�832.

Hojnik, J. and M. Ruzzier (2016). �What drives eco-innovation? A review of an

emerging literature.� Environmental Innovation and Societal Transitions 19:31�695

41.

Hyland, M. (2016). �Restructuring European electricity markets. A panel data anal-

ysis.� Utilities Policy, 38:33�42.

Jacobs J. P. A. M., J. E. Lighart and H. Vrijburg (2009). �Dynamic Panel Data Mod-

els Featuring Endogeneous Interaction and Spatially Correlated Errors.� Center700

Discussion Paper Series, 2009�92.

39

Jamasb, T. and M. Pollitt (2005). �Electricity market reform in the European Union:

review of progress toward liberalization and integration.� The Energy Journal,

26:11�41.

Jamasb, T. and M. Pollitt (2008). �Liberalisation and R&D in Network Industries:705

The Case of the Electricity Industry.� Research Policy, 37(6):995�1008.

Jobert, T., F. Karan�l and A. Tykhonenko (2018). �Degree of Stringency Matters:

Revisiting the Pollution Haven Hypothesis Based on Heterogeneous Panels and

Aggregate Data.� Macroeconomic Dynamics:1�23.

Joskow, P. L. (2006). �Markets for power in the United States: An interim assess-710

ment.� The Energy Journal, 27(1):1�36.

Joskow, P. L. (2008). �Lessons Learned from the Electricity Market Liberalization.�

The Energy Journal, 29(Special Issue 2):9�42.

Johnstone, N., I. Ha²£i£ and D. Popp (2010). �Renewable energy policies and techno-

logical innovation: evidence based on patent counts.� Environmental and Resource715

Economics, 45(1):133�155.

European Commission and UM Merit, (2007). �Final report MEI project about mea-

suring eco-innovation. Kemp, R. and P. Pearson.�.

Knittel, C. R. (2002). �Alternative regulatory methods and �rm e�ciency: Stochastic

frontier evidence from the US electricity industry.� The Review of Economics and720

Statistics, 84(3):530�540.

40

Krozer, Y. (2013). �Cost and bene�t of renewable energy in the European Union.�

Renewable Energy, 50:68�73.

Maddala, G. S., R. P. Trost, L. Hongyi and F. Joutz (1997). �Estimation of short-

run and long-run elasticities of energy demand from panel data using shrinkage725

estimators.� Journal of Business and Economic Statistics 15: 90�100.

Moreno, B., A. J. López and M. T. García-Álvarez (2012). �The electricity prices in

the European Union. The role of renewable energies and regulatory electric market

reforms.� Energy, 48(1):307�313.

Nakano, M. and S. Managi (2008). �Regulatory reforms and productivity: an empir-730

ical analysis of the Japanese electricity industry.� Energy Policy, 36(1):201�209.

Newbery, D. M. (1997). �Privatisation and liberalisation of network utilities.� Euro-

pean Economic Review, 41(3):357�383.

Pérez-Reyes, R. and B. Tovar (2009). �Measuring e�ciency and productivity change

(PTF) in the Peruvian electricity distribution companies after reforms.� Energy735

Policy, 37(6):2249�2261.

Pollit, M. (2008). �The arguments for and against ownership unbundling of energy

transmission networks.� Energy Policy, 36(2):704�713.

Pompei, F. (2013). �Heterogeneous e�ects of regulation on the e�ciency of the elec-

tricity industry across European.� Union countries.� Energy Economics, 40:569�740

585.

41

Ray S. C. and E. Delsy (1997) �Productivity Growth, Technical Progress, and E�-

ciency Change in Industrialized Countries� The American Economic Rewiev, 87:

1033�1039.

Ray S. C. �Data Envelopment Analysis: Theory and Techniques for Economics and745

Operations Research.� Cambridge University Press, 2004.

Roodman, D. (2009). �How to do xtabond2: An Introduction to Di�erence and

System GMM in Stata.� Stata Journal, 9(1):86�136.

Scheel, H. (2001): �Undesirable outputs in e�ciency valuations.� European Journal

of Operational Research, 132(2):400�410.750

Shehata, E. and S. Mickaiel (2013). �Spregdpd: Stata module to estimate Spatial

Panel Arellano-Bond Linear Dynamic Regression: Lag & Durbin Models.� Statis-

tical Software Components. https://ideas.repec.org/c/boc/bocode/s457506.html.

Simar, L. and P. Wilson (1999). �Estimating and bootstrapping Malmquist indices.�

European Journal of Operational Research, 115(3):459�471.755

Staiger, D. and J. H. Stock (1997) �Instrumental Variables Regression with Weak

Instruments.� Econometrica, 65(3):557�586.

Steiner, F. (2000). �Regulation, Industry Structure, and Performance in the Electric-

ity Supply Industry.� OECD Economics Department Working Papers, 238:1�39.

https://EconPapers.repec.org/RePEc:oec:ecoaaa:238-en.760

42

Stigka, E. K., J. A. Paravantis and G. K. Mihalakakou (2014). �Social acceptance of

renewable energy sources: A review of contingent valuation applications.� Renew-

able and Sustainable Energy Reviews, 32:100�106.

Teece, D., R. Rumelt, G. Dosi and S. Winter (1994). �Understanding corporate

coherence: Theory and evidence.� Journal of Economic Behavior & Organization,765

23(1):1�30.

Veugelers, R. (2012). �Which policy instruments to induce clean innovating?.� Re-

search Policy, 41(10):1770�1778.

Yang, H. and M. Pollit (2009). �Incorporating both undesirable outputs and uncon-

trollable variables into DEA: The performance of Chinese coal-�red power plants.�770

European Journal of Operational Research, 197(3):1095�1105.

43

Table and �gures

Table 1: Descriptive statistics of variables used in DEA analysis.

Variable Mean Std. Dev. Min Max Obs.

Electricity overall 4667.708 7160.487 1.3 26693.9 N = 162Generation between 7263.664 5.044 25219.78 n = 18

within 997.307 468.185 7853.685 T = 9

Electrical overall 21244.01 23498.89 439 87747 N = 162Capacity between 23991.31 451.111 77140.67 n = 18

within 1923.848 12820.35 31850.350 T = 9

Employment overall 49.31974 62.33176 .668 306.117 N = 162between 63.157 .769 283.813 n = 18within 9.168 3.486 104.126 T = 9

GGE overall 216.285 286.948 1.628 1358.756 N = 162between 272.647 1.839 994.612 n = 18within 107.244 -80.739 898.701 T = 9

Input overall 17562.5 21411.83 288.6 87610.8 N = 162Fuels between 21819.37 445.144 83783.84 n = 18

within 2187.636 8202.092 24572.3 T = 9

44

Table 2: Descriptive statistics for variables used in the econometric analysis.

Variable Mean Std. Dev. Min Max Obs.

E�ciency overall 0.997 0.076 0.777 1.443 N: 144between 0.020 0.953 1.050 n: 18within 0.073 0.747 1.413 T: 8

Pure e�ciency overall 0.998 0.065 0.799 1.347 N: 144between 0.023 0.963 1.079 n: 18within 0.061 0.814 1.294 T: 8

Scale overall 1.000 0.071 0.787 1.407 N: 144between 0.014 0.976 1.051 n: 18within 0.070 0.736 1.356 T: 8

TFP overall 1.026 0.095 0.797 1.657 N: 144between 0.047 0.978 1.147 n: 18within 0.083 0.676 1.536 T: 8

Regulation overall 2.239 0.811 0.871 4.434 N: 14between 0.705 1.016 3.445 n: 18within 0.430 1.480 3.764 T: 8

EPS overall 2.777 0.543 1.4 4.133 N: 14between 0.440 2.075 3.592 n: 18within 0.333 1.811 3.623 T: 8

R&D intensity overall 0.033 0.029 0.000 0.143 N: 14between 0.028 0.001 0.107 n: 18within 0.011 -0.01 0.070 T: 8

CHP penetration overall 0.166 0.136 0 0.786 N: 14between 0.125 0.030 0.462 n: 18within 0.059 -0.04 0.586 T: 8

pca fuelmix overall 0.020 1.179 -2.68 2.717 N: 14between 0.633 -1.30 1.052 n: 18within 1.004 -3.44 4.042 T: 8

45

Table 3: Average of the Malmquist Index and its main components.

Country Frontier Shift Ch. Pure E�. Ch. Scale E�. Ch. TFP Ch.

Austria 1.049 1 0.994 1.043Belgium 1.085 1.073 0.973 1.133Czech Republic 1.001 0.994 0.997 0.992Denmark 1.004 1 0.991 0.996Finland 1.005 0.982 0.997 0.984France 1.113 0.977 1.029 1.119Germany 0.996 1 1.002 0.998Greece 0.995 1.011 1.002 1.008Hungary 0.994 0.978 1.006 0.978Ireland 0.999 1 0.998 0.996Italy 1.052 1 1 1.052Netherlands 1.036 1 1 1.036Poland 1.046 1 1 1.046Portugal 1.037 0.978 1 1.015Slovakia 1.033 1.004 0.982 1.018Spain 1.026 0.996 1 1.023Sweden 1.061 0.96 0.988 1.006United Kingdom 0.99 0.986 1.012 0.988

mean 1.028 0.996 0.998 1.023a Note that all TFP index averages are geometric means.

46

Table4:Changein

totalfactorproductivityanditscomponentbetween2006-2014

Frontier

ShiftCh.(%)+

PE�.Ch.(%)+

ScaleE�.Ch.(%)=

TFPCh.(%)

Country

Frontier

ShiftCh.(%

)Pure

E�ciency

Ch.(%)

ScaleE�.Ch.(%)

TFPCh.(%)

Austria

2.1361

0-0.551

1.5567

Belgium

1.3104

0.3182

0.1177

1.7505

Czech

Republic

0.0572

-0.168

-0.014

-0.141

Denmark

-0.339

0-0.463

-0.803

Finland

-0.686

-1.061

0.6020

-1.140

France

1.0507

1.4459

3.0260

5.6340

Germany

0.1293

0-0.889

-0.767

Greece

-0.267

0.3637

0.7162

0.8310

Hungary

-0.114

4.0362

-3.008

0.8074

Ireland

-0.532

0-0.057

-0.604

Italy

-0.128

00

-0.128

Netherlands

-0.827

00

-0.827

Poland

2.8494

00

2.8494

Portugal

-0.294

-0.577

-0.137

-1.006

Slovakia

0.0853

2.3900

-2.178

0.2509

Spain

-0.527

0.1264

0.3182

-0.082

Sweden

2.1245

-2.443

0.2359

-0.139

United

Kingdom

0.0573

-0.379

0.0399

-0.284

47

Table5:Regulatione�ectfortheelectricitysectorontheannualMalm

quistindex.

(1)

(2)

(3)

(4)

(5)

(6)

TFP

b/se

b/se

b/se

b/se

b/se

b/se

L.TFP

-0.187***

-0.192***

-0.125***

-0.216***

-0.205***

-0.171***

0.024

0.024

0.043

0.026

0.024

0.041

Regulation

-0.032***

-0.035***

-0.038***

-0.031***

-0.025**

-0.029**

0.004

0.005

0.01

0.008

0.012

0.013

EPS

0.055***

0.026**

0.004

0.011

EPSMarket

0.022***

0.021***

0.019***

0.002

0.004

0.005

EPSNonmarket

0.031***

0.022**

0.012**

0.004

0.008

0.006

Tax

0.049***

0.013

TradingSchem

es0.027***

0.005

Feedin

Tari�s

0.005***

0.001

Standards

-0.005

0.012

R&Dsubsidies

0.017***

0.004

CHPpenetration

-0.325*

-0.158***

0.161

0.041

R&Dintensity

-0.798

-0.968

0.751

0.651

FUELMIX

0.008***

0.010***

0.003

0.003

N108

108

108

108

108

108

j23

2426

2526

27ar1p

0.093

0.091

0.057

0.109

0.089

0.061

ar2p

0.247

0.249

0.247

0.243

0.214

0.231

hansenp

0.695

0.703

0.801

0.804

0.738

0.791

48

Table6:Regulatione�ectfortheelectricitysector,ontheannualPure

E�ciency

change

(1)

(2)

(3)

(4)

(5)

(6)

Pe�

b/se

b/se

b/se

b/se

b/se

b/se

L.Pe�

-0.201***

-0.204***

-0.116***

-0.198***

-0.208***

-0.244***

00.004

0.02

0.005

0.005

0.025

Regulation

-0.007***

-0.007***

-0.007***

-0.006**

-0.007***

-0.007*

00.001

0.002

0.002

0.002

0.003

EPS

0.010***

-0.003

0.001

0.004

EPSMarket

0.001

0.001

-0.007

0.001

0.001

0.004

EPSNonmarket

0.008***

0.005**

0.005

0.002

0.002

0.006

Tax

0.024***

0.006

TradingSchem

es0.007***

0.002

Feedin

Tari�s

-0.012***

0.002

Standards

0.004

0.003

R&Dsubsidies

0.003*

0.001

CHPpenetration

-0.178***

-0.200***

0.01

0.014

R&Dintensity

0.871***

0.795***

0.067

0.114

FUELMIX

00

0.001

0.001

N108

108

108

108

108

108

j23

2426

2526

27ar1p

0.048

0.051

0.061

0.053

0.059

0.059

ar2p

0.281

0.305

0.483

0.3

0.218

0.224

hansenp

0.689

0.861

0.941

0.945

0.867

0.93

49

Table7:Regulatione�ectfortheelectricitysector,ontheannualScaleE�ciency

change

(1)

(2)

(3)

(4)

(5)

(6)

Scale

b/se

b/se

b/se

b/se

b/se

b/se

L.Scale

-0.312***

-0.357***

-0.390***

-0.354***

-0.291***

-0.281***

0.002

0.016

0.025

0.009

0.016

0.035

Regulation

-0.013***

-0.024***

-0.023***

-0.025***

-0.011***

-0.017***

0.003

0.004

0.007

0.004

0.004

0.006

EPS

0.004***

-0.001

0.001

0.003

EPSMarket

0.029***

0.029***

0.020**

0.004

0.004

0.008

EPSNonmarket

-0.028***

-0.030***

-0.041***

0.003

0.004

0.008

Tax

0.011

0.009

TradingSchem

es0.011**

0.005

Feedin

Tari�s

0.012***

0.002

Standards

-0.019***

0.005

R&Dsubsidies

-0.015***

0.002

CHPpenetration

-0.359***

-0.415***

0.015

0.021

R&Dintensity

-0.480***

1.105

0.114

0.996

FUELMIX

0.001

-0.002

0.002

0.002

N108

108

108

108

108

108

j23

2426

2526

27ar1p

0.09

0.085

0.111

0.087

0.075

0.01

ar2p

0.193

0.193

0.187

0.19

0.225

0.245

hansenp

0.775

0.702

0.96

0.62

0.864

0.965

50

Table8:Regulatione�ectfortheelectricitysector,ontheannualtechnologyinnovationchange

(1)

(2)

(3)

(4)

(5)

(6)

Tech

b/se

b/se

b/se

b/se

b/se

b/se

L.Tech

-0.018***

-0.012**

-0.059**

-0.082**

-0.015**

-0.068*

0.041

0.043

0.08

0.087

0.051

0.062

Regulation

-0.005

0.006

0.002

0.001

-0.011

00.004

0.009

0.01

0.007

0.01

0.012

EPS

-0.049***

-0.041***

0.007

0.01

EPSMarket

-0.067***

-0.065***

-0.065***

0.009

0.01

0.012

EPSNonmarket

0.012**

0.015

0.034**

0.004

0.009

0.016

Tax

-0.034***

0.01

TradingSchem

es-0.031***

0.005

Feedin

Tari�s

-0.004*

0.002

Standards

-0.004

0.009

R&Dsubsidies

0.005

0.004

CHPpenetration

-0.132*

-0.208**

0.075

0.092

R&Dintensity

-0.558

-1.401*

0.367

0.696

FUELMIX

0.004

0.005*

0.002

0.003

N108

108

108

108

108

108

j23

2426

2526

27ar1p

0.081

0.061

0.068

0.208

0.057

0.082

ar2p

0.926

0.785

0.989

0.959

0.977

0.8

hansenp

0.625

0.652

0.909

0.692

0.653

0.781

51

Table9:BayesianRegulatione�ectfortheelectricitysector,ontheannualTFPchange.

Mean

St.Dev.

MCSE

(Median

crlLow

ercrlUpper

∆TFPt

=∆TFPt−

1+

∆Regt+

∆EPSt+

∆v t

L.TFP

-0.180

0.0236

0.0008

-0.180

-0.226

-0.132

Regulation

-0.031

0.0043

0.0001

-0.031

-0.040

-0.022

EPS

0.054

0.0036

0.0001

0.054

0.047

0.061

∆TFPt

=∆TFPt−

1+

∆Regt+

∆EPSMarkett+

∆EPSNonmarkett+

∆v t

L.TFP

-0.194

0.0222

0.0017

-0.193

-0.237

-0.150

Regulation

-0.034

0.0046

0.0004

-0.034

-0.043

-0.025

EPSMarket

0.022

0.0018

0.0001

0.022

0.018

0.025

EPSNonmarket

0.030

0.0037

0.0002

-0.030

0.023

-0.038

∆TFPt

=∆TFPt−

1+

∆Regt+

∆EPSMarkett+

∆EPSNonmarkett+∑ K k=

1γkxik

+∆v t

L.TFP

-0.344

0.0261

0.0042

-0.343

-0.396

-0.295

Regulation

-0.004

0.0179

0.0036

-0.002

-0.044

0.0238

EPS

0.0106

0.0271

0.0065

0.0121

-0.041

0.0558

CHPpenetration

0.4544

0.2539

0.0533

0.4525

-0.090

0.8674

R&Dintensity

-29.33

0.4182

0.1198

-29.30

-30.06

-28.66

Fuelmix

-0.002

0.0038

0.0007

-0.002

-0.011

0.0041

52

Table10:BayesianRegulatione�ectfortheelectricitysector,ontheannualpure

e�ciency

change.

Mean

St.Dev.

MCSE

(Median

crlLow

ercrlUpper

∆Peff t

=∆Peff t−1

+∆Regt+

∆EPSt+

∆v t

L.Pe�

-0.200

0.000

2.928

-0.200

-0.201

-0.200

Regulation

-0.007

0.000

2.691

-0.007

-0.007

-0.005

EPS

0.009

0.000

7.034

0.009

0.008

0.010

∆Peff t

=∆Peff t−1

+∆Regt+

∆EPSMarkett+

∆EPSNonmarkett+

∆v t

L.Pe�

-0.203

0.002

0.000

-0.203

-0.208

0.199

Regulation

-0.006

0.000

7.077

-0.007

-0.008

-0.005

EPSMarket

0.000

0.000

4.023

0.000

-4.09

0.001

EPSNonmarket

0.007

0.001

8.687

0.007

0.005

0.009

∆Peff t

=∆Peff t−1

+∆Regt+

∆EPSMarkett+

∆EPSNonmarkett+∑ K k=

1γkxik

+∆v t

L.Pe�

-0.360

0.011

0.002

-0.360

-0.379

-0.338

Regulatin

-0.04

0.004

0.000

-0.04

-0.04

-0.03

EPS

0.07

0.015

0.004

0.06

0.05

-0.10

CHPpenetration

-0.27

0.019

0.004

-0.26

-0.31

-0.23

R&Dintensity

-2.83

0.212

0.060

-2.80

-3.24

-2.49

Fuelmix

0.007

0.001

0.000

0.007

0.004

0.010

53

Table11:BayesianRegulatione�ectfortheelectricitysector,ontheannualscalee�

ciency

change.

Mean

St.Dev.

MCSE

(Median

crlLow

ercrlUpper

∆Scalet

=∆Scalet−

1+

∆Regt+

∆EPSt+

∆v t

L.Scale

-0.311

0.0015

0.0001

-0.311

-0.314

-0.308

Regulation

-0.013

0.0023

0.0001

-0.013

-0.018

-0.009

EPS

0.003

0.0006

4.9174

0.003

0.002

0.004

∆Scalet

=∆Scalet−

1+

∆Regt+

∆EPSMarkett+

∆EPSNonmarkett+

∆v t

L.Scale

-0.352

0.0129

0.0018

-0.351

-0.376

-0.328

Regulation

-0.025

0.0029

0.0003

-0.025

-0.030

-0.019

EPSMarket

0.0289

0.0030

0.0003

0.0290

0.0228

0.0349

EPSNonmarket

0.028

0.0027

0.0003

0.028

0.022

0.033

∆Scalet

=∆Scalet−

1+

∆Regt+

∆EPSMarkett+

∆EPSNonmarkett+∑ K k=

1γkxik

+∆v t

L.Scale

-0.228

0.0199

0.0034

-0.226

-0.268

-0.194

Regulation

-0.014

0.0061

0.0012

-0.015

-0.023

0.0010

EPS

0.0082

0.0050

0.0010

0.0083

-0.001

0.0180

CHPpenetration

0.3958

0.0185

0.0025

0.3971

0.3546

0.4285

R&Dintensity

2.0500

0.3270

0.0915

2.0346

1.5494

2.6510

Fuelmix

-0.001

0.0023

0.0004

-0.001

-0.007

0.0017

54

Table12:BayesianRegulatione�ectfortheelectricitysector,ontheannualtechnologyinnovationchange.

Mean

St.Dev.

(MCSE

(Median

crlLow

ercrlUpper

∆Techt

=∆Techt−

1+

∆Regt+

∆EPSt+

∆v t

L.Tech

0.001

0.039

0.002

0.001

-0.07

0.079

Regulation

-0.00

0.004

0.000

-0.00

-0.01

0.002

EPS

-0.04

0.006

0.000

-0.04

-0.06

-0.03

∆Techt

=∆Techt−

1+

∆Regt+

∆EPSMarkett+

∆EPSNonmarkett+

∆v t

L.Tech

-0.015

0.0410

0.0035

-0.015

-0.094

0.0662

Regulation

0.0053

0.0078

0.0005

0.0053

-0.009

0.0211

EPSMarket

0.0114

0.0038

0.0002

0.0115

0.0039

0.0190

EPSNonmarket

-0.064

0.0082

0.0006

-0.064

-0.081

-0.048

∆Techt

=∆Techt−

1+

∆Regt+

∆EPSMarkett+

∆EPSNonmarkett+∑ K k=

1γkxik

+∆v t

L.Tech

-0.449

0.1245

0.0320

-0.446

-0.680

-0.233

Regulation

0.0472

0.0157

0.0033

0.0487

0.0169

0.0749

EPS

-0.018

0.0149

0.0032

-0.018

-0.047

0.0107

CHPpenetration

0.0615

0.1246

0.0283

0.0635

-0.172

0.2864

R&Dintensity

-9.263

0.4990

0.1331

-9.314

-10.07

-8.293

Fuelmix

0.0112

0.0039

0.0008

0.0111

0.0043

0.0189

55

Figure 1: Changes in sector and environmental regulation.

AustriaBelgium

Czech_RepublicDenmark

Finland

France

Germany

Greece

Hungary

Ireland

Italy

NetherlandsPoland

Portugal

Slovak

Spain

Sweden

United_Kingdom

11.

52

2.5

3M

arke

t Reg

ulat

ion

2013

1 2 3 4 5Market Regulation 2006

(a) Changes in the overall sector regulation.

Austria

BelgiumCzech_Republic

Denmark

FinlandFrance

Germany

Greece

Hungary

Ireland

Italy

Netherlands

Poland

Portugal

Slovak

Spain

Sweden

United_Kingdom

22.

53

3.5

4E

PS

201

3

1.5 2 2.5 3 3.5EPS 2006

(b) Changes in environmental policy strin-gency.

Figure 2: Country speci�c Malmquist index evolution by di�erent changes in the degree of sectorregulation.

.951

1.05

1.11.1

5tfp

ch_a

dd

2006 2008 2010 2012 2014Year

tfpch_add tfpch_addtfpch_add

Countries with increasing market regulation

(a) Countries with increasing sectorregulation.

.81

1.21.4

1.6tfp

ch_a

dd

2006 2008 2010 2012 2014Year

tfpch_add tfpch_addtfpch_add tfpch_addtfpch_add tfpch_add

High Persistently market regulation Cuontries

(b) Countries with persistently high sectorregulation.

.8.9

11.1

1.2tfp

ch_a

dd

2006 2008 2010 2012 2014Year

tfpch_add tfpch_addtfpch_add tfpch_addtfpch_add tfpch_addtfpch_add

Low Persistently market regulation Cuontries

(c) Countries with persistently low sectorregulation.

.91

1.11.2

1.3tfp

ch_a

dd

2006 2008 2010 2012 2014Year

tfpch_add tfpch_add

Countries with decreasing market regulation

(d) Countries with decreasing sectorregulation.

56

Figure 3: Country speci�c Malmquist index evolution by di�erent changes in the degree of envi-ronmental policy stringency.

.9.95

11.0

51.1

tfpch

_add

2006 2008 2010 2012 2014Year

tfpch_add tfpch_add

Countries with increasing EPS

(a) Countries with increasing EPS.

.81

1.21.4

1.6tfp

ch_a

dd

2006 2008 2010 2012 2014Year

tfpch_add tfpch_add tfpch_addtfpch_add tfpch_add tfpch_addtfpch_add tfpch_add

Countries with persistently high EPS

(b) Countries with persistently high EPS.

.91

1.11.2

1.3tfp

ch_a

dd

2006 2008 2010 2012 2014Year

tfpch_add tfpch_addtfpch_add tfpch_addtfpch_add

Low Persistently EPS Cuontries

(c) Countries with persistently low EPS.

.85.9

.951

1.05

1.1tfp

ch_a

dd

2006 2008 2010 2012 2014Year

tfpch_add tfpch_addtfpch_add

Countries with decreasing EPS

(d) Countries with decreasing EPS.

57

Figure 4: Country's Kernel density of market regulation stringency Bayes estimates.

0

20

40

60

80

-.05 -.04 -.03 -.02 -.01

overall1st-half2nd-half

Density of DFMalm_add_3:DOverall_3

(a) Czech Republic

0

20

40

60

80

100

-.05 -.04 -.03 -.02 -.01

overall1st-half2nd-half

Density of DFMalm_add_13:DOverall_13

(b) Poland

0

20

40

60

80

100

-.05 -.04 -.03 -.02 -.01

overall1st-half2nd-half

Density of DFMalm_add_1:DOverall_1

(c) Belgium

0

20

40

60

80

100

-.05 -.04 -.03 -.02 -.01

overall1st-half2nd-half

Density of DFMalm_add_14:DOverall_14

(d) United Kingdom

58

Figure 5: Country's Kernel density of environmental policy stringency Bayes estimates.

0

50

100

150

-.07 -.06 -.05 -.04

overall1st-half2nd-half

Density of DFMalm_add_18:DEPS_18

(a) United Kingdom

0

50

100

150

-.07 -.06 -.05 -.04

overall1st-half2nd-half

Density of DFMalm_add_1:DEPS_1

(b) Poland

0

50

100

150

-.07 -.06 -.05 -.04

overall1st-half2nd-half

Density of DFMalm_add_14:DEPS_14

(c) Portugal

0

50

100

150

-.065 -.06 -.055 -.05 -.045

overall1st-half2nd-half

Density of DFMalm_add_8:DEPS_8

(d) Greece

59

Appendix

This appendix demonstrates the link between MIC and MIV through the scale e�-

ciency term (Ray and Delsy, 1997).

MIC = MIV ×[σt(xj, yj)

t+1

σt(xj, yj)t× σt+1(xj, yj)

t+1

σt+1(xj, yj)t

]1/2

Recall MIV the product of the catch up e�ect and the frontier shift term:

MIV =δt+1V ((xj, yj)

t+1)

δtV ((xj, yj)t)×[δtV ((xj, yj)

t)

δt+1V ((xj, yj)t)

× δtV ((xj, yj)t+1)

δt+1V ((xj, yj)t+1)

]1/2

The scale e�ciency term expressed is the geometric mean of the two scale e�ciency

ratios:

[σt(xj, yj)

t+1

σt(xj, yj)t× σt+1(xj, yj)

t+1

σt+1(xj, yj)t

] 12

=

[δtC((xj , yj)

t+1)

δtV ((xj , yj)t+1)/δ

tC((xj , yj)

t)

δtV ((xj , yj)t)

] 12

×

[δt+1C ((xj , yj)

t+1)

δt+1V ((xj , yj)t+1)

/δt+1C ((xj , yj)

t)

δt+1V ((xj , yj)t)

] 12

=

[δtC((xj, yj)

t+1)

δtC((xj, yj)t)· δ

t+1C ((xj, yj)

t+1)

δt+1C ((xj, yj)t)

] 12

×[δtV ((xj, yj)

t)

δtV ((xj, yj)t+1)· δt+1

V ((xj, yj)t)

δt+1V ((xj, yj)t+1)

] 12

60

Multiplying the scale e�ciency term and the MIV we have the expression:

[δtC((xj, yj)

t+1)

δtC((xj, yj)t)· δ

t+1C ((xj, yj)

t+1)

δt+1C ((xj, yj)t)

] 12

×[δtV ((xj, yj)

t)

δtV ((xj, yj)t+1)· δt+1

V ((xj, yj)t)

δt+1V ((xj, yj)t+1)

] 12

δt+1V ((xj, yj)

t+1)

δtV ((xj, yj)t)×[δtV ((xj, yj)

t)

δt+1V ((xj, yj)t)

× δtV ((xj, yj)t+1)

δt+1V ((xj, yj)t+1)

]1/2

Simplifying only the �rst geometric mean in the squared brackets remains:

[δtC((xj, yj)

t+1)

δtC((xj, yj)t)· δ

t+1C ((xj, yj)

t+1)

δt+1C ((xj, yj)t)

] 12

that is the MI in CRS environment.

61