Determinants of the Efficiency of Electricity Generation in Latin America and Caribbean Countries Using a Cragg’s Regression Model

Abstract

Measuring and understanding the evolution of the electricity generation efficiency sector is a relevant task for policymakers. This study focuses on assessing the determinants of electricity generation efficiency across 24 countries in Latin America and the Caribbean between 2000 and 2020. A hurdle model was employed to account for the censored nature of the dependent variable. The findings show a positive impact of opening up the electricity sector in fully efficient countries. Market liberalization and the involvement of privately owned power plants contribute to efficiency gains in inefficient countries. Additionally, the tariff protectionism level exhibits positive effects for completely efficient and inefficient countries. These results provide valuable insights, as less-efficient countries can glean strategies from their more efficient counterparts to enhance their electricity generation efficiency.

1. Introduction

Trade between different markets and different products has positive effects on productive efficiency. One explanation of this relationship is the complementarity of pro-duction and demand patterns between the markets that take part in the exchange, for ex-ample, substituting the most expensive generation with the cheapest energy from imports. The benefit of the exchange derives from the scale of trade as well as the geographic scope of the interconnected markets [1].

In electricity generation, the potential gains from regional interconnection are associated with complementarities in the demand for electricity and the different endowments of primary energy resources among trading countries due to differences in seasonal generation and consumption patterns [2,3]. Interconnections can provide important benefits such as (a) security of supply [4,5], (b) efficiency in generation costs [3,6] (c) increased competition [3], and (d) a decrease in environmental costs [3].

The risk of generation system failure is referred to as supply security. In addition, the stochastic demand behaviour generates uncertainties in the required quantity, leading to a higher demand than the supply on some occasions. Faced with this situation and considering the inelasticity of electricity demand in the short term, regional interconnection increases supply security since electricity imports allow full demand coverage [4,5].

Different studies mention that interconnections allow an optimal allocation of re-sources for electricity generation, unlike autarky between countries, which induces excessive levels of generation capacity and unnecessary and risky investments in new plants that translate into high production costs [3,6]. In addition, the excessive capacity levels resulting from autarky in the electricity sector increase the price for the final consumer [4] and the volatility in prices in the different regions [7]. According to Mulder and Giesbertz [7], interconnection allows more consumers to access the electricity service because the marginal cost would otherwise exceed some consumers’ willingness to pay.

The increase in competition is evidenced in the integration between border markets, which promotes the entry of new companies with different sources of generation that minimize production costs and increase production efficiency. Finally, the decrease in environmental costs is caused by the increase in using new generation sources based on renewable resources such as wind and sun. In addition, the increase in the use of these clean sources reduces the challenges of balancing the energy system generated by their intermittent nature [3].

Despite the benefits of electricity interconnection, political and regulatory aspects largely determine its scope. Sometimes, the reforms create entry barriers for new competitors in regional markets. For example, price subsidies reduce incentives to expand investment in export-oriented electricity generation to neighbouring countries and limit the benefits of expanding transmission across borders [2].

Considering the foregoing, we aim to evaluate the determinants of electricity generation efficiency in 24 countries of Latin America and the Caribbean from 2000 to 2020, to identify relevant factors such as the openness of the electricity sector as a proxy of integration between countries. We use the model proposed by Cuadros et al. [8], who measured—from a dynamic DEA model using desirable and undesirable outputs—the efficiency of the sector. The novelty of this paper is focused not only on evaluating the factors that explain the electricity generation efficiency of Latin American countries, but also on identifying policy measures that can be carried on by these countries to improve their efficiency levels. From the methodology, we can distinguish between efficient and nonefficient countries in terms of the electricity generation activity. This helps us suggest policy measures that improve efficiency in inefficient countries by investing in the identified factors that impact their productivity, and establishing the factors have allowed efficient countries leverage their productive capacity.

Further material is divided into several parts. Thus, in Section 2, we present the major studies related to the determinants of electricity generation efficiency. In Section 3, we present the methodology used to estimate the efficiency model. In Section 4, we show the descriptive statistics of the variables used in the efficiency evaluation and present the model results. Finally, in Section 5, we present the conclusions.

2. Literature Review

This section focuses on the progress shown by relevant studies that have identified how intercountry integration affects electricity generation efficiency.

The existing literature has identified the positive effect of integration on efficiency in electricity generation. Regarding empirical works, the relationship has been measured in different contexts such as the European Union [9], Asia [10], South Africa [11], Central America [12], and South America, in particular, Chile, Argentina, and Peru [3]. However, the methodology used in these studies has been similar, focusing on the use of simulations that capture different scenarios.

Furthermore, integration can be associated not only with the augmentation of competition [3], but also with bolstering supply security [4,5] and enhancing efficiency in generation costs [3,6]. For example, Agostini et al. [3] demonstrates the viability of transparent and clear energy exchange between Chile and its neighbouring nations, thereby diminishing marginal energy costs and overall operational expenses, while maintaining a relatively consistent average generation cost. Moreover, Newbery and Grubb [4] showed that interconnectors facilitate the physical importation of power, providing supplementary alternatives to fulfil domestic demand. Consequently, it can be inferred that interconnectors generally contribute to the enhancement of supply security, except in exceedingly rare scenarios characterized by extreme combinations of contractual and physical factors. Finally, Brinkerink et al. [6], based on a literature review, identified pivotal trends associated with the global grid concept, revealing a notable decline in costs pertaining to long-distance transmission technologies.

Furthermore, DG ENER [9] estimated the value of integration among European Union countries under different scenarios from 2015 to 2030 ranged from EUR 12.5 to EUR 40 billion, compared to the dependence on national autarky. For South Asia, Timilsina and Toman [10], there are estimated potential savings of up to USD 226 billion in the provision of electricity due to crossborder electricity trade between 2015 and 2040. In South America, Agostini et al. [3] simulated a regulatory framework for electricity trade between Chile, Argentina, and Peru, showing annual benefits of USD 40.6 million for the three countries because of the decrease in generation costs. In Central America, Echevarria et al. [12] estimated the benefit of the Electric Interconnection System for Central American Countries (SIEPAC) at USD 1409 billion between 2019 and 2025 in a higher integration scenario in the region. Finally, in South Africa, ESMAP [11] estimated the savings from full integration with a single market and only one system operator, ranging from $100 million to $1.5 trillion per year between 1997 and 2002. In general, integration between countries leverages the complementarities between them.

Other authors have evaluated the factors that affect productivity in the electricity generation sector. For example, Ajayi [13] found that greater vertical integration associated with natural monopolistic market structures, a higher degree of public ownership, and higher barriers to entry, are statistically significant in increasing generation costs in OECD countries.

Plant ownership is a widely studied variable in the literature. Studies in different countries have found that privately owned plants tend to be more efficient than publicly owned ones because of their focus on making profits [14,15,16,17,18]. However, others have found that the difference in plant efficiency is greater for public plants or is not significant according to the type of property [19,20]; in the short run, Pollitt, [18].

The literature also investigates the diversification of matrix generation as a variable of interest. Specifically, the examination focuses on analysing the complementarity among resources, which subsequently influences strategic decision making concerning technological choices in electricity generation [21,22]. Studies show that investment in different sources of electricity generation can be beneficial by exploiting each country’s resources and taking advantage of geographical complementarities, reducing production costs, investment risks, and price volatility [21,22].

On the other hand, studies have focused on evaluating the impact of reforms in the electricity sector that seek to increase competition and encourage private investment in the sector. These studies identify the impacts of the reforms on installed capacity, labour productivity, and efficiency in electricity generation [23,24]. For example, Zhang et al. [23] found that competition among generators has positive effects on electricity generation, and electricity generation per worker is positively related to private participation within well-defined regulatory frameworks. In addition, Erdogdu [24] reveal that progress in the direction of electricity market reform is positively correlated with increased efficiency. Nevertheless, it is imperative to acknowledge that this effect is constrained and exhibits noteworthy variability across diverse country groups, evidencing the necessity to control at country-level. In this regard, the literature has substantiated that several variables, designed to control for country-specific effects, encompass factors such as industrialization, political rights, civil liberties, entry barriers, urbanization, climatic variables, electricity regulatory frameworks, and income [13,23,24,25]. In terms of tax of tariffs, according to Bustos [26], lower tariffs may induce foreign technology adoption.

3. Methodology

In this study, we analyse the determinants of efficiency in electricity generation for 24 countries in Latin America and the Caribbean for the 2000–2020 period. This section includes three subsections. The first contains the description of the data and the information. The second presents the measurements used to construct the variables. Finally, the third presents the model used to evaluate the relationship between integrating electricity markets and efficiency in the electricity generation sector proposed in this study.

3.1. Data and Information Sources

The data collected comes from different information sources. We obtained the electricity generation efficiency level from the previous work of Cuadros et al. [8]. Imports, exports, and electricity generation from the different generation sources come from the U.S. Energy Information Administration [27]. We obtained the average tariff rate level for all imported products from the World Bank. The number of firms with public and private capital to build private participation comes from Sielac-Olade [28].

Finally, we obtained the year in which the country liberalized the electricity sector based on different sources of information. In Argentina, we consulted Resolution 240 of 2003 [29]; in Bolivia, Law 1604 of 1994 [30]; in Brazil, Law 9648 of 1998 [31]; in Chile, we consulted the General Law of Electric Services [32]; in Colombia, we considered the information of the CREG’s website [33]; in the Dominican Republic, the Regulation for the application of the Electricity Law No. 125-01 [34]; in Ecuador, the Electrical Sector Regime Law [35]; in El Salvador, we took the information from the General Superintendence of Electricity and Telecommunications’ website [36]; in Guatemala, the National Electric Power Commission [37]; in Honduras, the General Law of the Electricity Industry [38]; in Mexico, we considered the information from the National Energy Control Center [39]; in Nicaragua, we consulted the Electricity Industry Law [40]; in Panama, we obtained the in-formation from Law No. 6 of the Regulatory and institutional framework for the provision of public electricity service [41]; in Peru, Decree Law 25844 [42]; and, in Uruguay, Law 16832 [43].

In sum,

Table 1. Variables and sources.

Variable	Description	Source
Dependent variable
Efficiency of electricity generation	Own calculation based on the Dynamic DEA model	Cuadros et al. [8]
Variable of interest
Openness	Own calculation: Share of electricity exports and imports over the total electricity generation	The U.S. Energy Information Administration
Control variables
Diversification of the energy matrix	Own calculation: Diversification of the generation matrix based on the entropy concept	The U.S. Energy Information Administration
Ownership of plants	Own calculation: Participation of privately owned power plants in the generation sector	Sielac-Olade
Market liberalization	Own calculation using information on a country-level basis	Different sources of information on a country-level basis
Tariff rate	Unweighted average of the rates applied for all products subject to tariffs calculated for all traded goods	World Bank

presents the description and sources of the variables.

3.2. Measurements

In this section we describe the dependent variable (efficiency in electricity generation), the variable of interest, and, finally, the control variables.

3.2.1. Dependent Variable

The dependent variable is the efficiency of electricity generation that we obtained from the work of Cuadros et al. [8], based on a dynamic Data Envelopment Analysis (DEA) model. The DEA was proposed in Charnes et al. [44] and is one of the most important tools for measuring relative efficiency in different fields [45].

The model involves the generation of electricity as a desirable output; as undesirable output, the CO₂ emissions produced in the generation activity; and, as inputs, the GDP per capita and the installed capacity in the different generation sources. This score ranges from 0 to 1, indicating the minimum and maximum efficiency, respectively.

The model is based on the dynamic Data Envelopment Analysis (DEA) model of Constant Returns to Scale (CRS) based on slacks proposed by Tone and Tsutsui [46], extended to include the undesirable interdependence between electricity generation of electricity from fossil sources and CO₂ emissions, presented in Zhou and Liu [47].

The DEA model defines a production possibility frontier for each year based on the outputs and inputs of each country. We used the dynamic DEA model proposed by Cuadros et al. [8], whose formulation is based on Zhou and Liu [47] and that asserts the maximization of generation of electricity, and the minimization of CO₂ emissions can be achieved from an additive model by the Equation (1):

$$\max SDO\_N{F}_{ot}+SDO\_F_{ot}+SUO\_C{O}_{2_{ot}}$$

(1)

where SDO_NF_ot and SDO_F_ot refer to the inefficiencies of generation of electricity from nonfossil and fossil sources, respectively, and SUO_CO_2ot is the excesses of the emissions caused by the electricity generation activity. In the DEA literature, these inefficiencies and excesses are known as slacks.

The production possibility set for each country (country o = 1, …,24) under CRS is defined by Equations (2)–(10).

Equations (2)–(4) represent the constraints on inputs:

$$GD{P}_{ot}={{\displaystyle \sum }}_{j=1}^{24}GD{P}_{jt}{\lambda }_j^t+S\_GD{P}_{ot}$$

(2)

where ${\lambda }_j^t$ is the linear combination coefficient of each country j in year t. The GDP of country “o” must be less than or equal to the linear combination of the GDP of all countries in each year t. The difference is the slack variable of the GDP of country o in year t (S_GDP).

$$IC\_N{F}_{ot}={{\displaystyle \sum }}_{j=1}^{24}IC\_N{F}_{jt}{\lambda }_j^t+SIC\_N{F}_{ot}$$

(3)

The nonfossil installed capacity (IC_NF) of country o must be less than or equal to the linear combination of the nonfossil installed capacity of all countries in each year t. The difference is the slack variable of the IC_NF of country o in year t (SIC_NF).

$$IC\_F_{ot}={{\displaystyle \sum }}_{j=1}^{24}IC\_F_{jt}{\lambda }_j^t+SIC\_F_{ot}$$

(4)

The fossil installed capacity (IC_F) of country o must be less than or equal to the linear combination of the fossil installed capacity of all countries in each year t. The difference is the slack variable of the IC_F of country o in year t (SIC_F).

Equations (5)–(6) represent the constraints of the electricity generation from nonfossil sources and fossil sources, respectively:

$$DO\_N{F}_{ot}={{\displaystyle \sum }}_{j=1}^{24}DO\_N{F}_{jt}{\lambda }_j^t+SDO\_N{F}_{ot}$$

(5)

The generation of electricity from nonfossil sources (DO_NF) of country o must be less than or equal to the linear combination of electricity generation from the nonfossil sources of all countries in each year t. The difference is the slack variable of the DO_NF of country o in year t (SDO_NF).

$$\alpha DO\_F_{ot}={{\displaystyle \sum }}_{j=1}^{24}DO\_F_{jt}{\lambda }_j^t+SDO\_F_{ot}$$

(6)

The electricity generation from fossil sources (DO_F) of country o must be less than or equal to the linear combination of electricity generation from the fossil sources of all countries in each year t. The difference is the slack variable of the DO_F of country o in year t (SDO_F). In Equation (6), α represents the interdependence between generation from fossil sources and the CO₂ emissions.

$$\alpha UO\_C{O}_{2ot}={{\displaystyle \sum }}_{j=1}^{24}UO\_C{O}_{2jt}{\lambda }_j^t+SUO\_C{O}_{2ot}$$

(7)

The CO₂ emissions from the fossil sources (UO_CO₂) of country o must be less than or equal to the linear combination of CO₂ emissions of all countries in each year t. A variation in electricity generation from fossil sources is accompanied by the same proportional variation in CO₂ emissions.

Following the model proposed by Cuadros et al. [8], the level of installed capacity for each country in each year t determines the installed capacity in the immediately succeeding year. Equations (8)–(9) represent this assumption:

$${{\displaystyle \sum }}_{j=1}^{24}IC\_N{F}_{jt}{\lambda }_j^t={{\displaystyle \sum }}_{j=1}^{24}IC\_N{F}_{jt}{\lambda }_j^{t+1}$$

(8)

$${{\displaystyle \sum }}_{j=1}^{24}IC\_F_{jt}{\lambda }_j^t={{\displaystyle \sum }}_{j=1}^{24}IC\_F_{jt}{\lambda }_j^{t+1}$$

(9)

Equation (10) captures the CRS assumption:

$${{\displaystyle \sum }}_{j=1}^{24}{\lambda }_j^{t+1}\ge 0$$

(10)

Additionally, non-negativity conditions are required:

$$S\_GD{P}_t, SDO\_N{F}_t, SDO\_F_t, SIC\_N{F}_t, SIC\_F_t, SUO\_CO{2}_t\ge 0$$

(11)

Finally, for each country, we calculated the measure for each year (${\tau }_{ot}^{*}$) and the overall efficiency measure (${\tau }_o^{*}$), from the Equations (12)–(13), respectively:

$${\tau }_{ot}^{*}=\frac{1}{1+\frac{1}{3}\left(\frac{SDO\_N{F}_{it}}{DO\_N{F}_{it}}+\frac{SDO\_F_{it}}{DO\_F_{it}}+\frac{SUO\_N{F}_{it}}{UO\_C{O}_{2it}}\right)};t=2000, \dots , 2020$$

(12)

$${\tau }_o^{*}=\frac{1}{21}{{\displaystyle \sum }}_{2000}^{2020}{\tau }_{ot}^{*}{}_{it};t=2000, \dots , 2020$$

(13)

a country o will be overall efficient ${\tau }_o^{*}=1$ if and only if is efficient in each year.

From Equation (12), a country with input or CO₂ emission excess will be an inefficient country, and will have an overall efficiency measure less than 1, given by the simple average of the efficiency measure year over year as shown in Equation (13).

In sum, the dynamic DEA model measures efficiency in generation of electricity for each country and year, accounting for inefficiencies in generation or excesses in inputs.

In the work of Cuadros et al. [8], they cover the 2000–2016 period. We used the same formulation of Dynamic DEA, we updated the data and we covered the period from 2000 to 2020.

3.2.2. Variable of Interest

The variable of interest is the level of openness in the electricity sector, which captures the integration level, and it is calculated based on the imports and exports level as a proportion of electricity generation. This variable is proposed considering that interconnections provide benefits as security of supply [4,5], reducing the generation costs [3,6], increasing competition [3], and decreasing environmental costs [3].

The following is the definition of the country-level trade openness metric:

$$\frac{x+m}{GDP}$$

(14)

where $x$ refers to exports, $m$ refers to electricity imports, and GDP is the gross domestic product.

However, to capture the openness of the generation sector, we propose to calculate the share of electricity exports and imports over the total electricity generation in country i in year t, following the mathematical expression:

$$Opennes{s}_{it}=\left(\frac{X_e{}_{it}+M_e{}_{it}}{E{G}_{it}}\right)\ast 100$$

(15)

where $X_e{}_{it}$ refers to electricity exports, $M_e{}_{it}$ represents the electricity imports, and $E{G}_{it}$ is the electricity generation in country i, year t.

Thus, a country in complete autarky in the electricity sector presents an openness level of zero, for example, island countries. Countries with relatively high volumes of electricity trade are those with the highest levels of openness.

3.2.3. Control Variables

For both industries and nations, we propose four control variables. We suggest diversifying the energy matrix and establishing a private ownership percentage for the companies engaged in generation activity at the industry level. Nation-wise, we use the liberalization of the electricity market and the tax rate. We explain each of the variables below.

• Diversification of the Energy Matrix

The literature has evaluated the benefit of diversification in terms of risk reduction and not as a competitiveness strategy. In this sense, Garcia-Mazo [21] analysed the diversification of the electricity generation matrix from financial and strategic points of view. Greater diversification represents a better response to the uncertainty caused by fluctuating fuel prices and climate, as well as a more efficient use of natural resources and a lower environmental. Consequently, investing in different generation sources with seasonal and geographic complementarities not only reduces investment risks, costs, and price volatility, but also increases the benefits of generation activity [22].

We carried the measurement of diversification in the generation sector, using the entropy index [48]. The Shannon entropy index used is derived from a previous measure as a measure of dispersion in a distribution [49].

We used the Shannon entropy index for country i in year t as a proxy for the diversification of the generation matrix and we calculated it from the following expression:

$$Diversificatio{n}_{it}=e^{-{{\displaystyle \sum }}_{j=1}^7{p}_{ijt}\ln p_{ijt}}$$

(16)

where $p_{ijt}$ represents the share of generation source of each of the seven sources j over the total generation of country i in year t. In our case, we the information disaggregated in seven generation sources: Nuclear, fossil fuels, hydroelectricity, geothermal, solar (or tide, wave, and fuel cell), wind, and biomass (and waste). Therefore, the value range for diversification is 1 to 7, where 1 denotes a country whose generation matrix is entirely concentrated in a single generation source and 7 denotes a country whose generation sources have the same participation in its generation matrix (1/7).

• Plant Ownership

Theoretically, competition in the electricity sector generates gains in terms of efficiency and productivity. This relationship has been evaluated in many empirical studies and most have found, controlling different factors, that privately owned plants are more efficient than publicly owned plants because they have better incentives to maximize profits rather than focusing on public services [14,15,16,17] and, in the long run, Pollit [18].

We calculated the participation of privately owned power plants in the generation sector in country $i$ from the following expression:

$$Privately owne{d}_i=\frac{\# private plant{s}_i}{\# total plant{s}_i}$$

(17)

where the numerator is the number of private plants in country i, at the end of the period, and the denominator is the total number of plants in country i, at the end of the period.

• Market Liberalization

State companies dominated the Latin American electricity systems before the reforms. These companies were vertically integrated for the generation, transmission, and distribution of electricity. Until the early 1990s, governments owned the electricity sector or controlled it through strict regulations. World reforms in the late 1980s aimed to break the idea of electricity as a natural monopoly, promoting competition and private investment in the sector [50].

We considered the year of market liberalization in the countries involved in the study. We used a dummy variable to indicate the year the electricity market in country i in year t was liberalized:

$$M_{it}=\left\{\begin{array}{c}1 if the electricity market is liberalized in the year t \hfill \\ 0 otherwise \end{array}\right.$$

(18)

• Tariff Rate

The simple average applied tariff was calculated as the unweighted average of the rates applied for all products subject to tariffs calculated for all traded goods. In this definition, tariff is a tax on goods and services imported and exported into a country. We used this measure as a proxy of willingness to adopt new technologies. Firms in countries with lower tariffs have incentives to import and adopt new technology from other countries.

The World Bank estimates this measure using the world integrated trade solution system from the World Trade Organization database.

3.3. Model Specifications

A hurdle model was used to identify the determinants of efficiency in electricity generation. This statistical model was used when the dependent variable was bounded on the left and on the right and showed an excess of concentration of the data over a lower or upper limit.

The hurdle model assumes that the censored observations are independent of the uncensored ones, and it was proposed by Cragg [51]. The model estimates two equations: one for the censored situation and another to explain the uncensored data. The model for censored data involved the analysis of the probability of attaining value 1 or 0 (binary model), and the model for uncensored data models, the probability of nonzero values (counting model). The model estimation was performed using maximum likelihood.

The mathematical formulation of the hurdle model in our study is as follows:

3.4. Binary Model: (Model of Efficient Countries)

$$s_i=\left\{\begin{array}{c}1 if z_i\gamma +u_i\ge 1\\ 0 if z_i\gamma +u_i<1\end{array}\right.$$

(19)

where $s_i$is a dummy variable that takes the value of 1 if the efficiency of electricity generation for country i is 1, and takes the value of 0 otherwise, $z_i$ is a vector of independent variables, γ is a vector of coefficients of independent variables, and $u_i$ is the error term distributed normally ($u_i~N\left(0,1\right)).$It is important to note that $s_i=1$ is estimated using a Probit model, whose mathematical formulation is as follows:

$$P(s_i=1)=z_i\gamma +u_i$$

(20)

Therefore, we estimate the variables that determine whether a country is efficient, using the Probit model proposed in (8).

3.5. Counting Model: (Model of Nonefficient Countries)

$$Y_i^{*}=x_i\beta +{\epsilon }_i$$

(21)

$$Y_i=Y_i^{*} if s_i=0 and Y_i^{*}>0$$

(22)

where $x_i$ is a vector of explanatory variables, β is a vector of coefficients, and ${\epsilon }_i~N\left(0,{\sigma }^2\right)$. It is estimated by using a truncated normal regression. $s_i$is a dummy variable that takes the value of 0 if the efficiency of electricity generation for country i is less than 1.

Therefore, we estimated the variables that determine whether a country presents efficiency of electricity generation less than 1, using the model formulated in (9).

In summary, we employed a two-step model known as a hurdle model. In the first model, we analysed the determinants of whether a country is efficient in electricity generation. In the second, we analysed the determinants of countries that are inefficient in electricity generation activity. Therefore, we model the efficiency of electricity generation in two steps.

Our data followed a panel structure. Due to the computational difficulty of including the panel data structure in hurdle models, we tested the temporal homoscedasticity hypothesis between countries that may affect the efficiency of regression coefficients estimated. The used test is a Z test and was contrasted against a normal distribution. The null hypothesis states that no variable affects the conditional variance in each model. In our case, we tested if the country variable affects the conditional variance in both the bi-nary and the counting models.

4. Results and Their Discussion

This section contains two parts. The first shows the descriptive statistics of the variables used in the proposed model. The second presents the results obtained from the estimated model.

4.1. Descriptive Analysis

Table 2. Descriptive statistics.

Country	Statistic		Dependent Variable	Independent Variables
		Efficiency Group	Efficiency	Openness	Diversification	Liberalization	Privately Owned	Tariff
AR	Mean	1	0.90	9.96	2.42	0.86	0.55	13.01
	SD		0.03	2.83	0.20	0.36		0.93
BO	Mean	2	0.83	0.06	2.10	1.00	0.58	10.39
	SD		0.07	0.12	0.13	0.00		1.39
BR	Mean	1	1.00	8.26	2.27	1.00	0.73	13.51
	SD		0.00	2.39	0.53	0.00		0.68
CH	Mean	1	0.93	1.75	2.62	1.00	0.7	6.28
	SD		0.04	1.74	0.45	0.00		0.81
CO	Mean	1	0.91	1.74	1.87	1.00	0.16	9.65
	SD		0.05	1.11	0.15	0.00		3.03
CR	Mean	3	0.33	6.85	2.27	0.00	0.18	5.74
	SD		0.24	3.25	0.30	0.00		0.33
CU	Mean	1	1.00	0.00	1.24	0.00	0	10.58
	SD		0.00	0.00	0.05	0.00		0.29
DR	Mean	2	0.75	0.00	1.54	0.90	0.32	8.34
	SD		0.11	0.00	0.14	0.30		2.51
EC	Mean	1	0.88	4.81	2.05	1.00	0.34	12.25
	SD		0.07	4.28	0.14	0.00		1.73
ES	Mean	2	0.74	13.23	3.52	1.00	0.06	6.32
	SD		0.10	10.33	0.55	0.00		0.58
GU	Mean	2	0.80	10.55	3.16	1.00	0.69	5.83
	SD		0.06	7.48	0.30	0.00		0.47
GY	Mean	3	0.68	0.00	1.29	0.00	0.23	11.59
	SD		0.08	0.00	0.05	0.00		2.30
HA	Mean	3	0.54	0.00	1.74	0.00	0	3.91
	SD		0.11	0.00	0.25	0.00		1.10
HO	Mean	2	0.83	4.29	2.80	0.33	0.74	5.86
	SD		0.07	3.51	0.98	0.48		0.48
JA	Mean	2	0.75	0.00	1.36	0.00	0.23	8.01
	SD		0.15	0.00	0.19	0.00		1.12
MX	Mean	1	1.00	1.15	2.09	0.24	0	11.65
	SD		0.00	0.99	0.13	0.44		4.51
NI	Mean	2	0.72	3.58	3.23	0.81	0.48	5.42
	SD		0.06	5.34	0.96	0.40		0.51
PN	Mean	2	0.81	2.78	2.12	0.57	0.2	7.20
	SD		0.09	1.40	0.23	0.51		0.70
PY	Mean	1	1.00	80.07	1.01	0.00	0	10.74
	SD		0.00	7.89	0.01	0.00		1.07
PE	Mean	1	0.96	0.05	2.06	1.00	0.49	6.70
	SD		0.09	0.08	0.28	0.00		3.97
SU	Mean	2	0.73	11.74	2.00	0.00	0	11.05
	SD		0.08	17.38	0.08	0.00		2.28
TT	Mean	3	0.51	0.00	1.01	0.00	0	8.98
	SD		0.27	0.00	0.01	0.00		1.30
UR	Mean	3	0.49	17.82	2.27	1.00	0.04	11.09
	SD		0.30	12.93	0.83	0.00		1.19
VE	Mean	1	0.99	0.68	1.87	0.00	0.09	13.14
	SD		0.02	0.36	0.08	0.00		0.53

Source: Own elaboration. Labels: AR: Argentina, BO: Bolivia, BR: Brazil, CH: Chile, CO: Colombia, CR: Costa Rica, CU: Cuba, DR: Dominican Rep., EC: Ecuador, ES: El Salvador, GU: Guatemala, GY: Guyana, HA: Haiti, HO: Honduras, JA: Jamaica, MX: Mexico, NI: Nicaragua, PN: Panama, PY: Paraguay, PE: Peru, SU: Suriname, TT: Trinidad and Tobago, UR: Uruguay, VE: Venezuela.

presents the descriptive statistics of the data set used. We classified the countries in three groups in relation to their overall efficiency levels.

In Table 2, the third column represents the efficiency group. We considered interesting to analyse the descriptive statistics for each of them using different maps for each variable.

Figure 1 shows the spatial distribution of the analysed variables using quantile maps.

Regarding efficiency, group 1 is constituted by 10 countries with overall efficiencies that vary between 0.87 and 1. These countries are: Brazil, Cuba, Mexico, Paraguay, Venezuela, Peru, Chile, Colombia, Argentina, and Ecuador. Cuba presents value of zero in terms of openness, which can be interpreted as that the electricity demand of this country is completely covered by its generation. This country presents the second lowest level of diversification of the countries of the group, and their power system does not have free competition, where all generation plants are publicly owned. In addition, countries like Venezuela, Mexico, Colombia, Chile, and Peru present values of openness higher than 0 and lesser than 1.75%, evidencing low energy exchanges with other countries. These countries stand out for being countries that diversify their generation with at least two sources. Of these countries, Venezuela is the only one that does not present free competition in the electricity market. On the other hand, Ecuador, Brazil, and Argentina present openness values between 4.8% and 10%, whose diversification is higher than 2, meaning that their generation is covered by at least three sources, and whose electricity markets are characterized by having been liberalized before the first year considered in the analysis period and since 2003 in the case of Argentina. Paraguay is the country with the highest level of openness, its exchange of energy with other countries is about 80%, but with a diversification level of 1, showing that the country’s generation matrix is almost absolutely oriented towards one only source: hydroelectricity. In addition, its electricity market has not been liberalized and all generation plants in the country are state-owned. Finally, the level of taxes applied to imported products in the countries of group 1 was 10.75% on average between 2000 and 2020. The highest tariff level was Brazil with 13.51%, while the minimum observed was Chile with 6.28%.

The second efficiency group is made up of nine countries and their overall efficiency levels vary between 0.71 and 0.84. These countries are: Bolivia, Honduras, Panama, Guatemala, Jamaica, Dominican Republic, El Salvador, Suriname, and Nicaragua. Of these countries, Jamaica, Dominican Republic, and Bolivia have null or almost null exchanges of electricity, since their openness values in the period analysed are 0 in the two first cases, and 0.06 in the Bolivian case. The range of the diversification measure of these countries is between 1.3 and 2.1 and implies that their generation involves two sources, in Jamaica and Dominican Republic and three sources in Bolivia.

In addition, Suriname and Jamaica are the only countries that do not present free competition in the electricity market. Regarding Suriname, it has the second largest ex-change of electricity in this group, with an 11.74% openness level.

Furthermore, Panama and Nicaragua have two aspects in common: (i) Their power systems were not subject to free competition from the beginning of the analysed period, and (ii) their plants are mostly publicly owned. In addition, the generation of electrical energy in Nicaragua depends on three sources, while in Panama it depends on two generation sources.

Finally, the bordering countries—Honduras, Guatemala, and El Salvador—showed openness levels between 4.29% and 13.33%. Moreover, their diversification indexes were between 2.8 and 3.6, implying that their generation matrices involve between three and four sources. In addition, Guatemala and El Salvador have had a free competitive power system before the analysed period, unlike Honduras, which only allowed free competition since 2014. Regarding plant ownership, almost all of them are publicly owned in El Salvador, while most are privately owned in Honduras and Guatemala. Lastly, the average level of taxes to imported goods in the nine countries was similar to the mean observed in 24 countries in the study, 8.91% compared with 9.19% for the 24 countries. The lowest tariff level was Nicaragua, with 5.42%, while the highest observed level was Suriname with an 11.05%.

The overall efficiency of the group of countries with the lowest generation efficiency levels varies between 0.32 and 0.68, consisting of Guyana, Haiti, Trinidad and Tobago, Uruguay, and Costa Rica. Of these countries, Haiti, Trinidad and Tobago, and Guyana have three things in common: (i) They do not have exchanges of electricity with other countries, showing values of zero in terms of openness, (ii) they use two sources of generation in their matrix, (iii) the electricity systems in these countries do not have free competition. On the other hand, all the plants of Haiti and Trinidad and Tobago are publicly owned, and, in Guyana, this participation is around the 77%. Regarding Uruguay and Costa Rica, they have openness levels of 17.82 and 6.85, respectively, being Uruguay the second largest extent of this index out of the 24 countries. Both countries generate electricity from three sources and their plants are mainly publicly owned. However, they differ because Uruguay has had a free competition electricity system before 2000 and Costa Rica does not have free competition in the analysed period. Finally, the level of taxes of imported goods is larger than 11.1% in Trinidad and Tobago, Uruguay, and smaller than 6% in Costa Rica, Guyana, and Haiti.

Below, we present the correlation matrix for the independent variables, which helps examine their relationship and detect potential multicollinearity problems in the estimations.

Table 3. Correlation matrix of independent variables.

Variable	Openness	Diversification	Private Ownership
Openness
Diversification	−0.1204
Private Ownership	−0.1948	0.4318
Tariff	0.1310	−0.3218	−0.1246

presents the correlation matrix of the independent variables. Due to all the correlations are less than 0.85, the independent variables do not present problems associated with multicollinearity [52,53].

4.2. Model Estimation

We used the hurdle model proposed by Cragg to identify the determinants of generation efficiency in 24 countries of Latin America and the Caribbean during the 2000–2020 period. It is important to note our data varied not only by country (i) but also by year (t).

We used linear and nonlinear factors to define the factors that determine the level of generation efficiency. We incorporated the linear factors into the analysis from the examined literature. The nonlinear factor was diversification because policymakers may be interested in identifying the inflection points that allow defining policies requiring transitions to these diversification levels.

In our study, the binary and counting models were given by:

Binary Model: (Efficient Countries Model)

Following the Probit model defined in (8), in our case, the binary model was given by:

$$P(s_{it}=1)={\gamma }_0+{\gamma }_{1}Opennes{s}_{it}+{\gamma }_{2}Diversification+{\gamma }_{3}Diversificatio{n}_{it}^2+{\gamma }_4{M}_{it}+{\gamma }_{5}PrivateOw{n}_{it}+{\gamma }_{6}Tarif{f}_{it}+u_{it}$$

(23)

where $s_{it}$ is a dummy variable that takes the value 1 whether country i in year t is efficient in its generation activity and takes the value of 0 otherwise, ${\gamma }_i$ is the vector of coefficients of independent variables, and $u_{it}$ is the error for country i in year t term distributed as a standard normal with mean zero and standard deviation 1.

Counting Model: (Nonefficient Countries Model)

Following (9) and (10), the counting model is defined as:

$$Y_{it}={\beta }_0+{\beta }_{1}Opennes{s}_{it}+{\beta }_{2}Diversification+{\beta }_{3}Diversificatio{n}_{it}^2+{\beta }_4{M}_{it}+{\beta }_{5}PrivateOw{n}_{it}+{\beta }_{6}Tarif{f}_{it}+{\epsilon }_{it}$$

(24)

where $Y_{it}$ represents the efficiency of electricity generation for country i in year t for countries whose efficiency is higher than 0 and less than 1, ${\beta }_i$ is a vector of coefficients of the independent variables, and ${\epsilon }_{it}$ is the error term for country i and year t, which is distributed as ${\epsilon }_{it}~N\left(0,{\sigma }^2\right)$. It is estimated by using a truncated normal regression.

Table 4. Binary and counting models.

Variable	Binary Model	Counting Model
	$\mathit{\gamma }$-Coefficients	$\mathit{\beta }$
Constant	1.057	−0.277
Openness	0.087 **	−0.051 *
Diversification	−8.750	1.909
Diversification2	1.431	−0.420
Market liberalization	−0.871	2.523 ***
Private Ownership	3.102	4.790 ***
Tariff	0.751 **	0.184
∑ Country	0.121 ***
Ω Country	0.008

p-values: *** 0.01, ** 0.05, * 0.1.

presents the hurdle model estimates. The left-hand side shows the estimates for the selection model, while the right-hand side contains the estimates for the uncensored or counting model.

Table 4 shows the estimation for the binary and counting models. We estimated the model fitting a heteroskedastic Probit using the identifier of the country for modelling the conditional variance in both the binary and counting models. We tested the null hypothesis of homoscedasticity between countries, and we found that this variable is only statistically significative for the Probit model.

We tested the null hypothesis of homoscedasticity between countries. ∑ represents the information about the estimated standard deviation of the error term in the selection model and Ω the information about the counting model. Results showed that the country variable is significant at 1% and there is evidence that the countries affect the variance of the selection model. On the other hand, there is no evidence that countries affect the variance of the counting model.

Regarding the results of the binary model, there is a higher tendency to be completely efficient when facing increases in openness in the electricity sector and increases in the generalized level of taxes in the economy, with significant effects occurring at the 5% level. Additionally, the type of property, the matrix diversification, and electricity market liberalization do not have a significant impact on the probability of observing an efficient result for a single year.

On the other hand, because the hurdle model is a nonlinear model, the coefficients of the counting model cannot be interpretable directly.

Table 5. Marginal effects.

Variable	dy/dx
Openness	−0.001
Diversification	−0.050
Diversification2	0.003
Market liberalization	0.104 **
Private Ownership	0.273 ***
Tariff	0.020 ***

p-values: *** 0.01, ** 0.05.

presents the marginal effects of each variable of the counting model.

The results show that, for countries with uncensored efficiency levels lower than 1, openness in the electricity sector does not have a significant impact on the level of efficiency, which is contrary to what was found in different regions around the world based on simulation of scenarios that capture a total integration between countries. For example, DG ENER [9] estimated the value of integration for the countries of the European Union between 2015 and 2030. Timilsina and Toman [10] did the same for the countries of South Asia between 2015 and 2040. In South America, Agostini et al. [3] simulated a regulatory framework for electricity trading between Chile, Argentina, and Peru. In Central America, Echevarria et al. [12] estimated the benefit of the SIEPAC between 2019 and 2025. Finally, in South Africa, ESMAP [11] estimated the savings from a complete integration, a single market and only one system operator between 1997 and 2002. All these studies pointed to important benefits of integration in terms of savings in electricity generation due to cross-border trade.

We did not find a significant impact of the diversification of the generation matrix on the efficiency of electricity generation. As far as we know, only Perez-Odeh [22] argued the existence of a positive relationship between diversification and efficiency.

The results suggest that the liberalization of the electricity market has positive and significant effects at 1% on efficiency. The liberalization of the market has an average impact of 0.104 on the efficiency level exhibited regarding efficiency in years when the market is not liberalized.

Private participation in the generation of electricity has positive and significant effects at 1%. Increases of 0.1 in the participation of private plants increase generation efficiency by 0.027 (0.273 × 0.1 = 0.027). This result follows the same path found in many studies that use the generation plant as the unit of analysis and find that those with private capital are more efficient than those with public capital in different countries. For example, See and Coelli [14] studied the efficiency of thermal plants in Malaysia between 1998 and 2005. Sarica and Or [15] included thermal, hydroelectric, and wind plants in Turkey. Khanna et al. [16] worked with fossil and thermal plants in India. Dean [17] analysed the efficiency of coal and natural gas plants in the United States, while Pollit [18] used different plants around the world, finding that long-term private plants are more efficient than public ones.

Finally, there is a positive and significant impact of tariff rate on efficiency. In other words, if the level of taxes in the economy increases 1%, the efficiency in the inefficient countries increases 0.02. Protecting local industry seems to have a positive impact on electricity generation efficiency in the region, contradicting Bustos’ result [26]. Olcay and Laing [54] found that governments can benefit and protect local industries from protectionist policies, like the pharmaceutical industry.

5. Conclusions

In this work, we evaluated the determinants of electricity generation efficiency in 24 Latin America and Caribbean countries during the 2000–2020 period using the proposed model by Cuadros et al. [8], derived from the DEA methodology. Our variable of interest measured the openness in the electricity sector as the share of exports and imports over the total generation in each country. The methodology chosen—based on the censored characteristic of the dependent variable—was the hurdle model proposed by Cragg [51]. This methodology allowed an independent analysis of fully efficient countries from those that showed some inefficiency in the analysed period.

From the methodology, we can separate the determinants of the efficiency of electricity generation for efficient countries and for the inefficient ones.

Regarding the efficient countries, the results suggest a positive and significant open-ness effect in the electricity sector in the sample, such as Brazil, Mexico, and Paraguay. The Cuba case is distinct since it lacks electricity interconnections.

For inefficient countries, we did not find evidence of a significant openness effect on efficiency. On the other hand, we found evidence that market liberalization had the strongest effect on efficiency gains. The liberalization of the market has an average impact of 0.104 on the efficiency level exhibited regarding efficiency in years when the market is not liberalized. Therefore, we consider that countries like Costa Rica, Guyana, Haiti, Jamaica, Suriname, and Trinidad and Tobago, could benefit from a liberalized electricity market in terms of greater energy efficiency. Following to Zhang et al. [23], inefficient countries can re-evaluate their regulatory electricity frameworks, taking the efficient ones as references.

In addition, we observed that participation of private plants has positive effects on efficiency for the inefficient countries. Increases of 0.1 in the participation of private plants increased generation efficiency by 0.023. In countries with low private ownership, such as Costa Rica, El Salvador, Guyana, Haiti, Jamaica, Panama, Suriname, Trinidad and Tobago, and Uruguay. We consider that this outcome is useful for policy makers to propose strategies focused on promotion of private investment in electricity generation sector.

Finally, the protectionism level of tariffs has positive effects for completely efficient and inefficient countries. Baldwin [55] proposes a theoretical model where the tariffs of the countries are noncooperative strategies, which leads to a suboptimal Nash equilibrium. The author found that unilateral tariff reductions were detrimental to the outcome of each country.

The evidence we present is important for the countries that are at the lowest level of efficiency—Haiti, Trinidad and Tobago, Guyana, Uruguay, and Costa Rica. Of these countries, Uruguay is the only one that has a liberalized electricity market. Moreover, all of them have low participation of private companies in electricity generation. We recommend that policymakers in these countries encourage private investment in the generation sector and implement policies aimed at liberalizing the electricity market. The regulatory frameworks of the most efficient countries provide elements that help improve their efficiency of electricity generation.

Further research could concentrate on evaluating joint reduction tariff policies and their effects on efficiency generation. Additionally, future research might aim to determine whether electricity prices in the region are converging. It could also assess the role of market integration in Latin American countries in shaping this trend.