WPS6920 Policy Research Working Paper 6920 Why Has Energy Efficiency Not Scaled-up in the Industrial and Commercial Sectors in Ukraine? An Empirical Analysis Gal Hochman Govinda R. Timilsina The World Bank Development Research Group Environment and Energy Team June 2014 Policy Research Working Paper 6920 Abstract Improvement of energy efficiency is one of the main higher costs of finance, and higher opportunity costs options to reduce energy demand and to reduce of energy efficiency investment are key barriers to the greenhouse gas emissions in Ukraine. However, large- adoption of energy efficiency measures in Ukraine. scale deployment of energy efficient technologies Institutional barriers particularly lack government has been constrained by several financial, technical, policies, which also contributes to the slow adoption of information, behavioral, and institutional barriers. This energy efficient technologies in the country. The results study assesses these barriers through a survey of 500 suggest targeted policy and credit enhancements could industrial and commercial firms throughout Ukraine. help trigger adoption of energy efficient measures. The The results from the survey were used in a cumulative empirical analysis shows strong inter-linkages among the multi-logit model to understand the importance of the barriers and finds heterogeneity between industrial and barriers. The analysis shows that financial barriers caused commercial sectors on the realization of the barriers. by high upfront costs of energy efficient technologies, This paper is a product of the Environment and Energy Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author may be contacted at gtimilsina@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Why Has Energy Efficiency Not Scaled-up in the Industrial and Commercial Sectors in Ukraine? An Empirical Analysis * Gal Hochman and Govinda R. Timilsina§ JEL Classification: Q4 Keywords: Energy efficiency, Ukraine, Barriers, Adoption, Discrete choice models, Cumulative Logit Model Sector: Energy *The authors woud like to thank Mook Bangalore, Michael Traux and Mike Toman for very helpful comments and suggestions. We also acknowledge the financial support of the World Bank’s Knowledge for Change Program (KCP). The views and interpretations are of authors and should not be attributed to the World Bank Group and the organizations they are affiliated with. § Hochman (gal.hochman@rutgers.edu) is an Associate Professor, Department of Economics, Rutgers University and Timilsina (gtimilsina@worldbank.org) is a Senior Research Economist, Development Research Group, World Bank. 1. Introduction The adoption of energy efficiency measures has been touted as a major policy option to curtail energy demand in response to increasing price volatility. Its importance has been further lauded to reduce greenhouse gas (GHG) emissions. The International Energy Agency (IEA) estimates that energy efficiency measures account for the highest potential of the total GHG mitigation required to limit global temperature rise by 2050 to 2°C above pre-industrial levels (IEA, 2012). Many studies that develop marginal abatement cost curves for GHG mitigation show energy efficiency measures entail negative costs (i.e., value of energy savings exceeds investment costs even if GHG mitigation benefits are not accounted for) and therefore these options are interpreted as ‘low hanging fruits’ for climate mitigation (McKinsey & Company, 2009; ESMAP, 2012; ADB, 1998). In practice, however, the scale of implementation of such seemingly win-win options is small relative to their apparent economic potential. The rationale for this disparity is that implementation of these options is constrained by financial, institutional, and information barriers (Jaffe and Stavins 1994; Howarth and Sanstad 1995; Sorell et al. 2004; Mundaca et al. 2013). Moreover, the economics of energy efficiency measures is normally evaluated using engineering benefit-cost approaches (e.g., Goldstein et al. 1990; Blumstein and Stoft 1995; Brown et al. 1998; McKinsey & Company 2009; Gillingham and Sweeney 2012) and such an analysis usually omits variables such as opportunity costs (Allcott and Greenstone, 2014). If the costs imposed by barriers are accounted for, energy efficiency measures would be expensive, and firms lose interest to adopt (Anderson and Newell, 2004). The gap between cost efficiency of energy efficiency measures and their implementation is also coined as the “energy efficiency gap” (Blumstein et al., 1980; DeCanio, 1993; Jaffe and Stavins 1994; Sanstad and Howarth, 1994; Schleich, 2009; Sorrell et al., 2004). A number of studies have attempted, through empirical analysis, to understand the energy efficiency barriers in different countries, economic sectors and energy end-uses (see e.g., Rohdin and Thollander, 2006; Sardianou, 2008; Schleich, 2009). Using semi-structured interviews of the largest 8 non-energy intensive manufacturing firms in Oskarshamn municipality in Sweden that had participated in government sponsored energy audits around 2000, Rohdin and Thollander 2 (2006) find that cost/risk of production disruption, other priorities not related to energy consumption, information search costs related to energy efficient appliances/devices, higher opportunity costs of investment, lack of sub-metering and split incentives with energy service companies all are barriers to adopt increased energy efficiency. Sardianou (2008) investigates the determinants of industrial decision-making with respect to energy efficiency investments in Greece through a survey of 779 industrial firms around 2005 followed by an empirical analysis using a Probit model. A majority (62%) of firms surveyed reported they did not consider energy saving a first priority although 52% of the sample reported energy saving as a decision criterion when installing new machines or buildings. Fifty-six percent of the sample reported that they would develop an energy conservation policy if a competitor industry had implemented relevant actions; while 70% of firms indicated they were not aware of existing new technologies. Based on a sample of 2,848 German commercial and services sector firms that were surveyed during the 1990s, Schleich (2009) econometrically assesses the relevance of various types of barriers to energy efficiency at the sectoral level and across fifteen sub-sectors. The analysis suggests the lack of information and priority-setting of upper management, who often do not consider energy efficiency as a strategic priority as the main barrier to energy efficiency improvement. Historically, energy consumption has remained inefficient in Ukraine due to ageing infrastructure and prolonged consumption subsidies (Ogaranko and Hubacek, 2013) These developments have strengthened calls for Ukraine to increase its clean energy base and improve its energy efficiency. The government has indeed enacted several policies to promote the adoption of clean and energy-efficient technologies (Trypolska, 2012) – however, the adoption of energy-efficient technologies remains slow. While barriers to clean energy adoption and options to improve investment have been examined by OECD (2012), to our knowledge, the same analysis has not been done for energy efficiency despite its strategic importance. It is therefore important to investigate the barriers to adoption of energy efficient technologies in the country. This study aims to empirically examine the energy efficiency barriers in Ukraine in the commercial and industrial establishments. Specifically, the study attempts to examine key questions related to energy efficiency barriers including: 3 • Size: do larger firms have greater incentives to invest in energy efficiency? • Energy in total production costs: do energy-intensive firms (i.e., firms with higher share of energy costs in total production costs) have greater incentives to invest in energy efficiency? • Ownership: are private firms more energy efficient than public firms? • Employment: are labor-intensive firms less energy efficient than capital-intensive ones? • Financing: have high upfront capital costs hindered the adoption of energy efficiency technologies? • Split incentives: are rented spaces where landlords pay energy bills less energy efficient than self-owned spaces? • Knowledge: is lack of knowledge about energy efficient technologies one of the key barriers? • Technical barriers: are there any technical barriers preventing scaling-up of energy efficiency measures? • Existing rules/regulations: have existing rules and regulations helped improve energy efficiency? • Firm’s bureaucracy: have a convoluted and complex internal decision process slowed down adoption of energy efficiency measures? The study employed a sample survey of 500 commercial/service and industrial establishments throughout the country done in 2012. The data collected were then used in a cumulative Logit model to estimate the importance of the various barriers to the adoption of energy efficiency measures in Ukraine. Our analysis shows financial barriers (e.g., high upfront costs or high costs of financing) are the key factors impeding firms’ investments in energy efficiency measures in Ukraine. Knowledge and technical barriers follow this. We find mixed results regarding split incentives, whereby the building is rented and/or jointly owned. Contrary to general intuition, the study does not find evidence to support the energy-intensity hypothesis that assumes energy-intensive firms are more likely to adopt energy efficient technologies compared to non-energy-intensive firms. Instead, firms with higher revenue per unit of energy consumption tend to invest more on energy efficiency improvements. The separation of industrial and commercial firms allow us to introduce heterogeneity among sectors that lead us to 4 suggest that the commercial sector, which includes the public sector, is less likely to invest in energy efficiency measures in the absence of policy interventions. The paper is organized as follows. The next section (Section 2) briefly discusses the methodology used to derive the results. Section 3 presents the data used in the analysis, while Section 4 presents preliminary results. The main results are presented in Section 5, and their robustness assessed in Section 6. We offer concluding remarks in Section 7. 2. The methodology To investigate the hypotheses specified in the introduction and to better understand the barriers to the adoption of cost-effective energy efficiency technologies in Ukraine, we estimate a discrete choice model using the sample of Ukrainian firms that participated in our survey. We choose a discrete choice model to estimate the factors influencing the adoption of energy efficiency measures. We first estimate a binary choice model. The dependent variable is a binary variable with possible values of invested/did not invest in energy efficiency measures in the past five years. The covariate vector includes several factors that in theory facilitate the adoption of energy efficiency measures or keep investments in such measures at bay. The log-likelihood function of our binary choice model is of the standard type. We look at the log-likelihood function due to convenience, while noting that the natural logarithm of the likelihood function is a monotonic transformation of the likelihood function, and that the log- likelihood function achieves its maximum value at the same point as the likelihood function itself. Let xi denote the vector of independent variables and let yi denote a dummy that equals 1 if the firm invested in energy efficiency technologies in the past five years and 0 otherwise. An observation is denoted with i and there are N observations: Where F is the cumulative distribution function associated with either a logit or a probit specification and β denotes the parameters estimated. This formulation results in the following log-likelihood function: 5 Our covariate vector includes firm revenues, share of energy cost in total production cost, firm privately owned, and several measures of potential barriers to the adoption of energy efficiency measures. Specifically, we use the survey data collected to construct the independent variables capturing the barriers to adoption of energy efficiency measures. The survey section used to collect this data focuses on the perceived barriers to adoption.. The questions of that section in the questionnaire quantify firms’ perceived barriers to the adoption of energy efficiency measures. We grouped the barriers into seven categories: financial, split, informational, technical, existing rules and regulation, low energy prices, and firm’s bureaucracy. Each category included questions pertaining to specific barriers in the category, where each question asked the respondent answering the questionnaire to quantify the importance of a specific barrier on the scale from 0 (no influence) to 3 (strong influence). The binary choice model discussed above quantifies the factors that affect firms’ decision whether to invest in energy efficiency measures. However, because of the ordinal response of our questionnaire where the response implicitly captures ever-increasing levels of investment in energy efficiency measures, we also employed the cumulative logit model. These models are defined for the probability of having an ordinal response that is less than or equal to the value R, relative to the probability of having a response greater than the value R: Where for an ordinal variable with 7 categories, 6 cumulative logit functions are defined. Each of these cumulative logit functions includes a “cutpoint” (i.e., its own intercept), , but all of the cumulative logit functions share the same set of parameters for the k predictors, i.e., . Note that the number of estimated parameters is significantly lower than that of a multinomial logit model, and is equal to (R-1)+k, as opposed to (R-1)*(k+1) that are required in a multinomial logit model. 6 This framework suggests that the following transformation estimates the cumulative probability, denoted , that a given response y is less than or equal to the ordinal category k: The estimated cumulative probability, denoted , is just the difference in the estimates’ cumulative probability between response category k and (k-1): where . DeMaris (2004) identified conditions under which linear regression treatment of ordinal response leads to robust analysis. These include more than 5 levels, large sample, and a response distribution that is not highly skewed across the ordinal range. Although once introducing into the calculations missing observations the sample is not too large and our sample does not meet all of these conditions, we elected to keep the linear regression and add it to our analysis. The linear regression treats the ordinal response as a continuous variable, and the estimated results are used to assess the robustness of the estimates of the cumulative Logit model. 3. The survey and data processing The study employed a sample survey of 500 commercial/service and industrial establishments throughout the country. To map the sample back to an unbiased representation of the survey population we weighted the survey data using the prevalence of different firms in the overall economy for each sample observation. The respondents rated each barrier on a scale from 0 to 3 based on its perceived influence on the implementation of energy efficiency measures to the firm. If a barrier’s influence was strong, it was given 3 points; 2 points were given for considerable influence and 1 point for little influence. If a barrier had no influence, it was given 0 points, and if it was not applicable for a 7 firm, it was marked as “No answer/Not applicable”. The specific questions pertaining to the barriers of each group are included in Table 1 below. Table 1: Specific questions asked to analyze energy efficiency barriers Financial barriers High upfront costs: Are upfront capital costs of energy efficient appliances and devices high? Lack of capital: Do financial institutions (Banks and other financial institutions) perceive energy efficiency investment as risky and therefore charge high premium? Low opportunity costs: Are there other priorities for capital investment, which can produce high returns? Low opportunity costs of appliances to be replaced: Is there any resale value of the replaced appliances, which still has a long operational life? Long payback period: Is payback period of efficient appliances/devices too long to discourage their implementation? Split incentives Bills paid by landlord: No incentives for the firm to reduce energy consumption as energy bills are paid by building/facility owners Split bills: No incentives for the firm to reduce energy consumption as energy bills are split among the building/facility tenants Knowledge, information and experience Metering: Lack of gas, electric and heat metering Awareness: Lack of awareness of the availability and/or benefits of deploying energy efficient processes and devices Information: Difficulties with obtaining necessary information Confidence: Lack of confidence on energy efficient devices and processes (they do not deliver the services at the level their promoters advocate) Experience: Lack of experience in energy efficiency measures Technical barriers Skilled personnel: Lack of skilled personnel to handle the efficient devices and processes Supplies: Lack of local supplies for equipment parts and very expensive purchasing from abroad, as well as long lead time to get equipment parts Reconfiguration: Installation of energy efficiency measures needs substantial reconfiguration of production process Malfunction and poor performance: Higher probability of malfunction or poor performance thereby disrupting production process Existing Rules and Regulation Government permits: Need to obtain government permits to deploy energy efficient devices and processes Property rights: Lack of legal protection of property rights Policy instruments: Administrative price setting, subsidies and cross subsidies Government policy: Lack of effective government policies to facilitate energy efficiency programs Unofficial payments: Unofficial payments demanded to receive government permits Institutional barriers Decision chain: Long decision chain on the firm The future: Uncertainty about the firm’s future Conflict of interest: Conflict of interests inside the firm Economic barriers Low priority: Low priority of the firm to reduce energy consumption; energy cost is not a big component of production costs due to low energy prices 8 To this set of variables, we also introduced variables that capture firm characteristics. These are (i) revenues, (ii) ownership structure (public or private), (iii) share of energy costs to total production costs, (iv) number of employees and (v) facility rented or owned. Before estimating the binary choice model, we reduced the dimensionality of the model. There are 25 questions assessing the importance of the different barriers to the adoption of energy efficiency measures. In addition, there are 5 variables that capture firms’ characteristics. There are another 6 variables when estimating the cumulative logit model. On the other hand, because of missing observations, in some of the runs we ended up with only 98 observations. We had too many variables. We therefore reduced the number of variables/factors using factor 2 analysis tools. Specifically, we used principle-component analysis. While using principle- component analysis we managed to reduce the number of parameters estimated from 36 to about 12, and thus increased precision when estimating the various factors affecting the adoption of energy efficiency measures. We employ these techniques to aggregate the various independent variables and compute the common factors. The eigenvalue is proportional to the portion of the sum of the squared distances of the points from their multidimensional mean. The principle-component analysis essentially rotates the set of points around their mean in order to align with the principal components. This moves as much of the variance as possible (using an orthogonal transformation) into the first few dimensions. The values in the remaining dimensions tend to be small and may be dropped with minimal loss of information. To this end, we use the rule of thumb that requires the eigenvalue to be greater than 1 for the factor to be included in the empirical analysis. The common factors were then used to aggregate the specific barriers to those mentioned in Section 4 and that are used in the regression analysis. The results of the principle component analysis are depicted in Appendix A. 2 Principal component analysis employs orthogonal transformation to convert observations of correlated variables (variables that belong to a certain group – e.g., financial barriers) into a set of values of linearly uncorrelated variables that are called principal components. It is used in macroeconomics to aggregate multi-dimension indicators and to clean the noise from observed series in the panel, which is poorly correlated with the rest of the panel (e.g., Avesani et al. 2006, and Forni et al. 2000). It has also being applied to complex dataset, which included multiple indicators, to construct social capital indices (Sabatini, 2005). For more on the asymptotic characteristics of factor analysis, see Bai (2003). 9 In sum, we substantially reduced the dimensionality of the data (from about 36 to 12 variables). We obtained one common factor with eigenvalue greater than 1 for each of the barriers analyzed bellow, as illustrated in Table 1A to 6A in Appendix A. Using the loading factors we computed the aggregate level explanatory variable and tested the importance of financial (hypothesis vi); knowledge (hypothesis vii); technical (hypothesis viii); whether existing rules and regulations (hypothesis ix) serve as barriers to the adoption of energy efficiency measures in Ukraine; the barriers created by the internal structure of the firm (hypothesis x); and energy prices (hypothesis xi). When assessing the barriers to the adoption of energy efficiency measures, we also included a seventh barrier: energy prices. This set of explanatory variables was then augmented as follows: 1. With variables that capture the size of the firm and are used to test hypothesis (i). 2. To test hypothesis (ii) we included in the regression the share of energy cost to the firm relative to total production costs. 3. A dummy variable that equals 1 if the facility is rented and 0 otherwise is used to test the split incentive hypothesis (hypothesis (iii)). 4. An ownership dummy that equals one if the firm is privately owned, and zero otherwise (i.e., public or foreign owned), is introduced into the analysis. The parameter is used to evaluate hypothesis (iv). 5. An employment variable is introduced to assess hypothesis (v). 4. The empirical analysis The key data used in the empirical analysis are summarized in Table 2 – barriers include financial, split, knowledge information and experience, technical, existing rules and regulations, institutional, and economic barriers. Overall, factors are normalized, such that the lowest value assigned to a barrier is zero. When modeling firm characteristics, we have one categorical (total revenues), one dummy variable (private ownership), and one continuous variable (share of energy costs). 10 Table 2. Data summary Variable Obs Mean Std. Dev. Min Max Firm characteristics Invest 389 0.7609254 0.4270676 0 1 Total revenues 334 2.730539 1.63827 1 8 Share of energy costs 316 11.43358 10.21259 0.3 70 Facility rented 499 1.817635 0.3865323 1 2 Private ownership 491 0.694501 0.4610882 0 1 Employment 233 2.613734 1.375957 0 6 Barriers to the adoption of energy efficiency measures Financial barriers 356 6.583883 2.884173 0 14.59367 Split Barriers 382 1.025171 1.562384 0 5.50279 Knowledge barriers 433 3.667689 3.445591 0 12.38596 Technical barriers 420 3.758199 2.553517 0 11.82016 Existing laws and regulation 296 4.832675 2.878283 0 10.71962 Internal institutions 401 0.6719663 0.6634264 0 2.093352 Energy prices 443 0.9006772 1.039524 0 3 We present the Pearson correlation coefficients among the various factors in Table 3. Although principle-component analysis controls for correlation within groups of variables, we wanted to better understand correlation between groups of variables. Some of the correlations suggest we should add an interaction term among the factors, which we do below. But before studying the importance of the interaction terms, we focus on our baseline model, which is without the interaction term. Table 3. The Pearson correlation coefficient Financial Split Knowledge Technical Existing rules Internal Energy Column1 barriers Barriers Barriers Barriers and regulations institutions prices Financial barriers 1.00 Split Barriers 0.30 1.00 Knowledge Barriers 0.48 0.57 1.00 Technical Barriers 0.62 0.29 0.59 1.00 Existing rules and regulations 0.46 0.56 0.60 0.54 1.00 Internal institutions 0.46 0.42 0.49 0.50 0.55 1.00 11 Energy prices 0.31 0.54 0.57 0.46 0.60 0.52 1.00 We begin with a linear binary model that investigates and evaluates the factors that might impede firms from making any investment in energy efficiency measures. Recall that the dependent variable in our binary model receives a value of 1 if the firm invested in energy efficiency technologies in the past 5 years and 0 otherwise. The parameters estimated are depicted in Table 4, where we depict both the Probit and the Logit model. Although the fit of the models is not very good, the outcome does suggest financial barriers are the key obstacles to the adoption of energy efficiency measures (i.e., we cannot reject hypothesis vi at a 5% significant level). Table 4. The binary model Column1 Column2 Column3 Variable Probit Logit Total revenues 0.0258 -0.0380 Share of energy cost 0.0246 0.0529 Private owned -1.0039 -2.3658 Financial factor -0.1186** -0.1892** Split factor -0.0203 -0.1294 Knowledge and information factor 0.0040 0.0380 Technical factor 0.2083 0.3800 Existing rules and regulation factor 0.0259 0.0652 Energy prices 0.0902 0.0020 Constant 1.2752*** 2.7960* N 98 98 F 6.1630 6.343536844 Legend: * p<0.10; ** p<0.05; *** p<0.01 We now present the baseline cumulative logit model, where we try and better understand the factors that guide firms in Ukraine when deciding if and how much to invest in energy efficiency measures. The results are depicted in Table 5, where the cumulative logit model is 12 depicted in addition to the linear regression model that assumes the investment decisions are a continuous variable. For most of the results we find similar results across the cumulative logit and the linear models, except for the technical barriers, which are significant under the linear model at a 10%, level but not significant under the cumulative logit model. While the F-statistic of the cumulative Logit model is 3028.94, it is less than 100 for the linear model. Thus, in what follows we focus on the cumulative logit model. Financial Barriers The analysis suggests financial barriers are the key barriers not only when contemplating whether to make an investment in energy efficiency measures, but also when firms decide how much to invest. While reviewing the literature on energy efficiency barriers in the industrial sector, Worrell (2009) also finds similar results. Table 5. The baseline model Column1 Column2 Column3 Variable Cumulative Logit Linear Log of revenues 2.4257*** 1.7974*** Log of energy cost share 0.1728 0.0146 Private owned -0.4782* -0.3466* Log of financial factor -0.9205** -0.6200** Log of split factor -0.2368 -0.1860 Log of knowledge and information factor -0.9285 -0.5954 Log of technical factor 1.3632 0.8870* Log of existing rules and regulation factor 0.2303 0.1157 Log of energy prices 0.4577* 0.3768* Constant -0.0494 Cutoff value Constant: cut 1 0.5802 Constant: cut 2 2.8455* Constant: cut 3 4.2898** Constant: cut 4 5.4113*** Constant: cut 5 5.6107*** Constant: cut 6 6.7458*** 13 Statistics N 98 98 F 3028.94 302.96 legend: * p<0.10; ** p<0.05; *** p<0.01 Moreover, with our dataset, we are able to split the observations into industrial and commercial firms to examine barriers specific to each sector. The barriers are ranked in the industrial sector as follows (we report in parenthesis the rank score that respondents put on the questionnaire): high upfront capital costs that are needed to invest in energy efficient appliances and devices (2.1), lack of capital (1.9), long payback period (1.8), low opportunity costs (1.4) and small monetary value of the replaced appliances (1.2). This ranking is illustrated in Figure 1. Similarly, the commercial firms rank the barriers as follows: high upfront capital costs (1.9), lack of capital (1.8), long payback period (1.5), low opportunity cost (1.4) and small monetary value of the replaced appliances (1.2). Overall, the industrial sector ranks various financial barriers higher, although the differences are not large. While analyzing conservation tax credits of the early 1980s in the U.S., Carpenter & Chester (1984) found that although 86% of those surveyed knew about the credit, only 35% used it, and of those firms that used it, 94% of investments made into energy efficiency would have been done regardless of the financial incentives (e.g. in the absence of policy). In other words, Carpenter & Chester (1984) do not find the role of financial barriers in inhibiting energy efficiency investments. Our findings contradict those reported in Carpenter & Chester (1984). Our cumulative logit model suggests that at the mean, an increase of 1 in the log of the financial barriers results in an increase of the probability that a firm not invest in energy efficiency measures by 0.92. The linear model finds a similar, yet smaller impact: an increase of 1 in the log of the financial barriers results in the investment variable declining by 0.62. Split barriers Split barriers combine two barriers that show the influence of splitting the responsibility of using energy resources with another side: “No incentives for the firm to reduce energy consumption as energy bills are paid by building/facility owners” and “No incentives for the firm 14 to reduce energy consumption as energy bills are split among the building/facility tenants.” Our baseline analysis rejects hypothesis (iii). It rejects the hypothesis that imperfect information yields underinvestment in energy efficiency measures. We return to this hypothesis below, where interaction terms among the various factors are introduced. Technical barriers, and knowledge and information barriers Technical barriers are the other major barriers. The installation of energy efficiency measures needs substantial reconfiguration of production processes, and lack of local supplies for equipment parts and very expensive purchasing from abroad are seen as most important technical barriers (1.5). Besides, industrial producers in Ukraine do not trust new devices: they name high probability of their malfunction or poor performance, which can result in disrupting production process as an important barrier (1.1). Commercial firms also report having little experience in energy efficiency measures (the factor having 1 point on average) and say they lack local supplies of equipment parts that are too expensive to purchase from abroad (1.1 on average). Otherwise, respondents of commercial sector did not indicate significant informational or technical barriers. Although the linear regression outcome suggests technical barriers impact adoption of energy efficiency measures, the baseline cumulative Logit model does not hold this claim (i.e., while the linear model cannot reject hypothesis viii at a 10% significant level, the cumulative logit model does reject this hypothesis). We also do not find support for information barriers (hypothesis vii) under the baseline analysis. Existing Rules and Regulations A lack of effective government policies to facilitate energy efficiency programs ranks highest among rules and regulation factors with an average assessment of 2.2 points (Figure 1). Most important rules and regulation barriers for commercial firms are the lack of effective government policies to facilitate energy efficiency programs (1.8), need to obtain government permits to introduce energy efficient devices and processes (1.5), and long decision chain on the firm (1.3). However, once firm characteristics are controlled, existing rules and regulations do not seem to result in large barriers to the adoption of energy efficiency measures. That is, we do not find support for hypothesis ix. 15 Energy Prices On the other hand, energy prices do play an important role, as predicted by theoretical models and documented in other work that investigated various economies and focused on durable goods (Gillingham et al., 2009) – hypothesis xi. Hughes (1991) suggests that the energy sector is very important to Eastern European economies for two reasons: Eastern European countries have higher energy prices than countries with equivalent levels of income but are also some of the most energy intensive economies in the world. Our results suggest that high-energy prices might be affecting firms’ demand for cost-effective energy efficiency measures. Energy Costs On average, the firms surveyed reveal that they would decrease their energy costs by one- third, if there were no barriers to energy efficiency measures. There is almost no difference between estimates of different sectors. Industrial firms expect to reduce their energy consumption costs by 35.8% on average, and commercial firms say that if not for the barriers, their energy expenditures would be lower by 38.3%. Firm Revenues We cannot reject hypothesis (i): size matters. Firms with higher revenues are more likely to invest in energy efficiency measures. In the sample population, industrial firms have more revenues. Further, there is no correlation between firms that earn more revenues and financial barriers (the pairwise correlation coefficient equals -0.04). Private Ownership Hypothesis (ii) is rejected: Privately owned firms in Ukraine invest less in energy efficiency measures. While in our sample population industrial firms make on average more revenues than commercial firms, relatively more industrial firms are privately owned. 16 17 Figure 1. Rating of barriers for energy efficiency by industrial and commercial firms Industrial/commercial investment levels While using the cumulative logit model and the estimated cutoff values, we calculated the predicted probability that a firm not invest (0), make a small investment (<20k), make a slightly larger investment (20k-100k), make an average investment (100k-500k), make a larger investment (500k-1M), make a large investment (1M-10M), or make a very large investment 18 (>10M). We depict the predicted probabilities while separating between industrial and commercial firms. Our analysis suggests that while 24% of commercial firms will not invest in energy efficiency measures, only 15% of industrial firms will not invest. Further, the calculated predicted probabilities suggest that industrial firms are more likely to have larger investments in energy efficiency measures than commercial firms. This conclusion is interesting given that, on average, commercial firms are more energy intensive: while the average share of energy costs in total production costs is slightly larger for commercial firms (19.9% versus 17.7%, respectively), industrial firms are more likely to invest in energy efficiency measures. Figure 2. Predicted probabilities of investment using the baseline model Industrial firms are more likely to be privately owned, while commercial firms are more likely to be publicly owned (Table 6). Our baseline model suggests that, on average, privately owned firms are less likely to invest in energy efficiency measures (Table 5). Table 6. Ownership % private % public % foreign % Total Commercial 59 38 4 100 Industrial 81 12 7 100 19 Interactions How do the various barriers interact? And what is their impact on the adoption of energy efficiency measures? While focusing on the interaction among the various barriers whose correlation coefficient is larger than 0.5, we investigate, for example, the impact of lack of knowledge on firm’s perception of technical barriers to the adoption of energy efficiency measures. Although we suspected interaction terms might convey new information, we could not add them to the baseline specification because of data limitations. Therefore, and to investigate the importance of the various interactions, we dropped the private owned firm dummy variable and included an interaction term, one at a time. The various models were estimated assuming a cumulative logit model, and the results are depicted in Appendix B. In Table 7 we present the model that has the greatest explanatory power – its F-Statistic is more than 1,000,000. Introducing other interaction terms resulted in models with substantially lower explanatory power. Table 7. Baseline model with an interaction of knowledge and technical barriers Variable Model I Log of revenues 2.741325190*** Log of energy cost share 0.130654024 Log of financial factor -0.915981282*** Log of split factor -0.235681133 Log of knowledge and information factor -2.100629047** Log of technical factor 0.394984634 Log of existing regulation factor 0.200068708 Log of energy cost 0.162249862 Knowledge * technical 0.913821119** Cutoff parameters omitted Statistics N 98 F 1.04E+06 20 Although we do observe some fluctuation in the coefficient values, overall, the significance of firms’ revenues and financial barriers is maintained among the various specifications (see Appendix B). Further, as long as we do not introduce an interaction term, which interacts with the financial barriers, the magnitude of the estimated revenue and financial barriers parameters remains relatively stable. When introducing various interaction terms, one at a time, we get mixed results with respect to split, knowledge and information, and technical barriers. The coefficient of these parameters is significant under some of the specifications modeled in Appendix B but not others. The introduction of an interaction term between information and technical barriers suggests that less informed firms (higher information barrier) results in firms underestimating the importance of the technical barrier (Table 7). In Appendix B Model VI we also depict a model that shows knowledge affects the impact of split incentives on adoption of energy efficiency measures and reduces the negative impact split incentives have on the amount invested in these measures. We also computed the predicted probability, when an interaction between knowledge and technical barriers is introduced into the empirical analysis. Introducing an interaction term skewed the predicted probabilities of the commercial firms’ investment patterns toward more investment but yielded less investment for the industrial firms. However, the results still suggest it is more likely to observe investment in energy efficiency measures by industrial firms than commercial firms. How do the results change when we address the missing data problem? Do we gain information when imputing data or does it just introduce noise and affect the precision of the parameters estimated? The robustness analysis below explores these questions. 5. Robustness In the main analysis we introduced weights to compensate for potential biases. We now further investigate ways of correcting for the missing data while evaluating the benefits of imputing data. Because of the large portions of data missing, as well as the absence of questions 21 answered by all respondents that can be used to impute the data, poor results were obtained when using the multiple imputation models. However, two simple imputations proved useful in further understanding our results. For the first we substituted missing observations pertaining to firms’ perception regarding barriers to the adoption of energy efficiency measures with 0, while for the second we substituted it with the mean value. Overall, the results supported our main findings although the size of the parameters estimated did change (see Appendix C). 6. Concluding remarks This study examines energy efficiency barriers to the industrial and commercial (including public) sectors in Ukraine by conducting a survey of 500 firms throughout the country. The results from the survey are then used in empirical (i.e., Logit and Probit) models to understand the importance of various barriers to the adoption of energy efficiency. The study finds that financial barriers, such as higher upfront investment costs of energy efficiency technologies, lack of capital and long pay-back period are the strongest barriers to the deployment of energy efficiency technologies in the both industrial and commercial sectors in Ukraine. Lack of effective government policies and existing regulation such as government permits required for the adoption of energy efficiency technologies are other key barriers. Predicted probabilities estimated by our study suggest that industrial firms are more likely to have larger investments than commercial firms despite the fact that the latter have, on average, slightly higher share of energy costs in the total production costs. Our study also suggests that energy price rises would yield more adoption of energy efficiency measures, as would the introduction of credit enhancement instruments. Although the study suggests policy that reduces upfront costs and risk will result in more adoption of energy efficiency measures in Ukraine, the analysis also suggests heterogeneity among firms and sectors. That is, our analysis finds differences in levels of investment in energy efficiency measures among sectors, with industrial firms investing more. This raises the question whether sectoral heterogeneity be accounted for while designing energy efficiency policy instruments. We plan to further investigate this in future research. 22 References 1. Allcott, H, and M. Greenstone. 2012. Is there an energy efficiency gap? MIT working paper 12-03 2. Anderson, S. T. and R. G. Newell. 2004. “Information Programs for Technology Adoption: The Case of Energy-Efficiency Audits.” Resource and Energy Economics. 26 (1): 27–50. 3. Asian Development Bank (1998), ALGAS Summary Report, ADB, Manila. 4. Avesani, R. G., A. G. Pascual, and J Li. 2006. A New Risk Indicator and Stress Testing Tool: A Multifactor Nth-to-Default CDS Basket. IMF Working Paper WP/06/105 5. Bai, J. 2003. Inferential Theory for Factor Models of Large Dimensions. Econometrica. 71 (1): 135–171. 6. Blumstein, C., and S. E. Stoft. 1995. Technical Efficiency, Production Functions and Conservation Supply Curves. Energy Policy. 23 (9): 765–68. 7. Blumstein, C., B. Kreig, L. Schipper, and C. York. 1980. Overcoming social and institutional barriers to energy efficiency. Energy. 5:355–72 8. Brown, Marilyn A., Mark D. Levine, Joseph P. Romm, Arthur H. Rosenfeld, and Jonathan G. Koomey. 1998. “Engineering–Economic Studies of Energy Technologies to Reduce Greenhouse Gas Emissions: Opportunities and Challenges.” Annual Review of Energy and the Environment 23, 287–385. 9. Carpenter EH, Chester TS. 1984. Are federal energy tax credits effective? A Western United States survey. Energy J. 5:139–49 10. DeCanio SJ. 1993. Barriers within firms to energy-efficient investments. Energy Policy 21:906– 14 11. DeMaris, A. 2004. Regression with social data. John Wiley & Sons: New York 12. Energy Sector Management Assistance Program. 2012. Planning for a Low Carbon Future: Lessons Learned from Seven Country Studies, World Bank, Washington, DC. 13. Forni, M., M. Hallin, M. Lippi, and L. Reichlin. 2000. Coincident and leading indicators for the EURO area. The Review of Economics and Statistics. 82 (82): 540–554. 14. Gillingham, K., R. G. Newell, K. Palmer. 2009. Energy Efficiency Economics and Policy. NBER Working Paper 15031. http://www.nber.org/papers/w15031 23 15. Gillingham, K., and J Sweeney. (2012). Barriers to implementing low-carbon technologies. Climate Change Economics. 3(4): 1-21. DOI: 1250019 (2012) [21 pages] DOI: 10.1142/S2010007812500194 16. Goldstein, D., R. Mowris, B. Davis, and K. Dolan. 1990. Initiating Least-Cost Planning in California: Preliminary Methodology and Analysis, Natural Resources Defense Council and the Sierra Club, prepared for the California Energy Commission Docket No. 88-ER-8, Revised May 10. 17. Howarth RB, Sanstad AH. 1995. Discount rates and energy efficiency. Contemp. Econ. Policy 13:101–9 18. Hughes, G. 1991. The Energy Sector and Problems of Energy Policy in Eastern Europe. Oxford Review of Economic Policy. 7(2): 77-99. 19. International Energy Agency. 2012. Energy Technology Perspectives 2012: Pathways to a Clean Energy System. IEA, Paris. 20. Jaffe, A., and R. Stavins. 1994. The Energy Paradox and the Diffusion of Conservation Technology. Resource and Energy Economics. 16(2): 91–122. 21. Mundaca, L., M. Mansoz, L. Neij and G.R. Timilsina. 2013. Transaction Costs Analysis of Low-Carbon Technologies, Climate Policy, Vol.13, No. 4, pp. 490-513. 22. McKinsey & Company (2009), Pathways to a Low Carbon Economy: Version 2 of the Global Greenhouse Gas Abatement Cost Curve. McKinsey & Company. http://www.mckinsey.com/clientservice/ccsi/pathways_low_carbon_economy.asp. 23. Ogaranko, L. and K. Hubacek. 2013. Eliminating Indirect Energy Subsidies in Ukraine: Estimation of Environmental and Socioeconomic Effects Using Input–Output Modeling. Journal of Economic Structure 24. Rohdin, P. and P. Thollander. 2006. Barriers to and driving forces for energy efficiency in the non-energy intensive manufacturing industry in Sweden. Energy. 31 (12): 1836– 1844. 25. Sabatini, F. 2005. The Role of Social Capital in Ecnomi Development: Investigating the Casual Nexus through Structural Equations Models. 26. Sanstad AH, Howarth RB. 1994. Consumer rationality and energy efficiency. Presented at Proc. Summer Study Energy Effic. Build., Berkeley, CA 27. Sardianou, E. 2008. Barriers to industrial energy efficiency investments in Greece, Journal of Cleaner Production, Vol. 16 (13): 1416–1423. 28. Schleich, J. (2009). Barriers to energy efficiency: A comparison across the German commercial and services sector, Ecological Economics, Vol. 68 (7): 2150–2159 24 29. Sorrell S., O'Malley, E., Schleich, J., Scott, S. 2004. The Economics of Energy Efficiency - Barriers to Cost-Effective Investment, Edward Elgar, Cheltenham.. 30. Trypolska, G. 2012. Feed-in tariff in Ukraine: the only driver of renewables’ industry growth? Energy Policy. 45; 645-653 31. Worrell, E., L. Bernstein, J. Roy, L. Price, and J. Harnisch. 2009. Industrial energy efficiency and climate change mitigation. Energy Efficiency. 2(2): 109-123. 25 Appendix A: In computing the common factor attributed to “financial” barriers to the adoption of energy efficiency measures, we obtained the eigenvalues of the various relevant factors. The analysis suggests we retain only one factor (Table 1A-a). We also obtained the extracted sum of the squared loading (Table 1A-b). We use the loading coefficients to calculate the financial barrier factor that we employ in the empirical analysis. The loading coefficients and eigenvalues enabled us to transition from five variables to one that explains more than 53% of the financial barriers data variability. Table 1A-a. Principle component analysis of financial barriers Factor analysis/correlation Number of obs 356 Method: Principal-component Factors retained 1 Rotation: (unrotated) Number of parameters 5 Factor Eigenvalue Difference Proportion Cumulative Factor1 2.68267 1.97536 0.5365 0.5365 Factor2 0.70732 0.03498 0.1415 0.678 Factor3 0.67234 0.13459 0.1345 0.8125 Factor4 0.53776 0.13785 0.1076 0.92 Factor5 0.39991 . 0.08 1 LR test: independent vs. saturated chi2(10)= 457.19 Table 1A-b. Factor loading (pattern matrix) and unique variance for financial barriers Variable Factor1 Uniqueness High Up front costs 0.7184 0.484 Lack of Capital 0.7952 0.3677 Low opportunity cost 0.7205 0.4809 Zero or very small monetary value 0.723 0.4773 Long pay back period 0.7018 0.5074 26 Next, we computed the eigenvalue of the split incentives (Table 2A-a). The eigenvalues suggest one common factor, with the loading factors depicted in Table 2A-b. This resulted in moving from two variables to one, but note that the single factor used explains more than 80% of the variability in the split barriers data. Table 2A-a. Principle component analysis of split barriers Factor analysis/correlation Number of obs 382 Method: Principal-component Factors retained 1 Rotation: (unrotated) Number of parameters 1 Factor Eigenvalue Difference Proportion Cumulative Factor1 1.68226 1.36452 0.8411 0.8411 Factor2 0.31774 . 0.1589 1 LR test: independent vs. saturated chi2(10)= 238.34 Table 2A-b. Factor loading (pattern matrix) and unique variance for split barriers Variable Factor1 Uniqueness Energy bills paid by building/facility owner 0.9171 0.1589 Energy bills shared among building/facility owner and firm 0.9171 0.1589 Next, we compute the eigenvalue of the information and knowledge incentives (Table 3A-a). The eigenvalues suggest one common factor, with the loading factors depicted in Table 3A-b. Here we reduced the number of variables in the empirical analysis from five to one, while the the factor chosen explains more than 68% of the variability in the data. Table 3A-a. Principle component analysis of information and knowledge barriers Factor analysis/correlation Number of obs 433 27 Method: principal-component Factors retained 1 Rotation: (unrotated) Number of parameters 5 Factor Eigenvalue Difference Proportion Cumulative Factor1 3.41871 2.84219 0.6837 0.6837 Factor2 0.57652 0.13668 0.1153 0.799 Factor3 0.43985 0.08195 0.088 0.887 Factor4 0.35789 0.15086 0.0716 0.9586 Factor5 0.20703 . 0.0414 1 LR test: independent vs. saturated chi2(10)= 1181.82 Table 3A-b. Factor loading (pattern matrix) and unique variance for split barriers Variable Factor1 Uniqueness No metering 0.7696 0.4077 Lack of awareness 0.8895 0.2088 Difficulty obtaining information 0.8623 0.2564 Lack of confidence in these measures 0.8003 0.3596 Lack of experience 0.8069 0.3489 We computed the eigenvalue of the technical barriers (Table 4A-a). The eigenvalues suggest one common factor, with the loading factors depicted in Table 4A-b. When focusing on the technical barriers, the single factor explains more than 57% of the variability. Table 4A-a. Principle component analysis of technical barriers Factor analysis/correlation Number of obs 420 Method: principal-component factors retained 1 Rotation: (unrotated) Number of parameters 4 Factor Eigenvalue Difference Proportion Cumulative Factor1 2.31472 1.56538 0.5787 0.5787 Factor2 0.74934 0.2142 0.1873 0.766 28 Factor3 0.53514 0.13434 0.1338 0.8998 Factor4 0.4008 . 0.1002 1 LR test: independent vs. saturated chi2(10)= 413.15 Table 4A-b. Factor loading (pattern matrix) and unique variance for technical barriers Variable Factor1 Uniqueness Skilled labor 0.7026 0.5064 Expensive imports and lack of domestic supply 0.7938 0.3699 Requires substantial changes to the production process 0.8116 0.3413 High probability of malfunction 0.7296 0.4676 When computing the eigenvalue of questions pertaining rules and regulations (Table 5A- a), we retain one common factor, with the loading factors depicted in Table 5A-b. We used the rule of thumb, that requires the eigenvalue to be greater than 1 for the factor to be included in the empirical analysis. Although the second factor is close to one, its contribution to explaining the variability is much smaller than the first factor (0.5153 versus 0.182). We therefore elected to use only the first factor in our empirical analysis. Table 5A-a. Principle component analysis of existing rules and regulations Factor analysis/correlation Number of obs 296 Method: principal-component factors retained 1 Rotation: (unrotated) Number of parameters 5 Factor Eigenvalue Difference Proportion Cumulative 29 Factor1 2.57632 1.66628 0.5153 0.5153 Factor2 0.91004 0.31833 0.182 0.6973 Factor3 0.59171 0.07042 0.1183 0.8156 Factor4 0.52128 0.12064 0.1043 0.9199 Factor5 0.40065 . 0.0801 1 LR test: independent vs. saturated chi2(10)= 363.58 Table 5A-b. Factor loading (pattern matrix) and unique variance for existing rules and regulations Variable Factor1 Uniqueness Government permits required 0.7873 0.3802 Lack of property rights protection 0.7467 0.4425 Administrative price setting 0.7174 0.4853 Government policy not effective 0.5878 0.6544 Unofficial payments demanded 0.7341 0.4612 Finally, we compute the eigenvalue of the internal barriers, i.e., the firms’ administration and bureaucratic barriers to the adoption of energy efficiency measures (Table 6A-a). The eigenvalues suggest one common factor, with the loading factors depicted in Table 6A-b. The common factor explains more than 62% of the variability. Table 6A-a. Principle component analysis of firm’s administrative barriers 30 Factor analysis/correlation Number of obs 401 Method: principal-component factors retained 1 Rotation: (unrotated) Number of parameters 3 Factor Eigenvalue Difference Proportion Cumulative Factor1 1.88518 1.19138 0.6284 0.6284 Factor2 0.69381 0.27279 0.2313 0.8597 Factor3 0.42101 . 0.1403 1 LR test: independent vs. saturated chi2(10)= 238.16 Table 6A-b. Factor loading (pattern matrix) and unique variance for firm’s administrative barriers Variable Factor1 Uniqueness Long decision chains 0.7044 0.5039 Uncertainty about firm's future 0.8139 0.3375 Conflict of interest inside the firm 0.8524 0.2734 Appendix B: Table 1B. Baseline model with an interaction term Variable Model I Model II Model III Log of revenues 2.741325190*** 2.182619969*** 2.465261230*** Log of energy cost share 0.130654024 0.24513135 0.169066623 Log of financial factor -0.915981282*** -1.148505292** -0.969937005** Log of split factor -0.235681133 0.176517929 -0.111932736 Log of knowledge and information factor -2.100629047** -1.313373421 -0.818492276 Log of technical factor 0.394984634 1.627786066* 1.327466515 Log of existing regulation factor 0.200068708 0.087423073 0.182224142 Log of energy cost 0.162249862 0.743714209** 0.623607336 Knowledge * technical 0.913821119** Knowledge * regulation -0.188530449 knowledge*energy price -0.119880911 regulation*technical regulation*energy price 31 Cutoff parameters omitted Statistics N 98 91 98 F 1.04E+06 38.62230397 7.65E+03 Variable Model IV Model V Log of revenues 2.100542807*** 2.028502915*** Log of energy cost share 0.270295821 0.236193307 Log of financial factor-1.149360445** -1.208104546** Log of split factor 0.180178453 0.400469767* Log of knowledge and information factor -1.443025467 -1.398589357 Log of technical factor 1.811733302 1.786825604* Log of existing regulation factor 0.062758406 0.172567756 Log of energy cost 0.755600114** 1.375906460** Knowledge * technical Knowledge * regulation knowledge*energy price regulation*technical -0.058099965 regulation*energy price -0.964637983 Cutoff parameters omitted Statistics N 91 91 F 16.96402005 18.93379284 Variable Model VI Model VII Log of revenues 2.431694291*** 2.393070223*** Log of energy cost share 0.19165519 0.320244444 Log of financial factor -1.018054681** -1.228787952*** Log of split factor -3.135320904*** -0.786425154 Log of knowledge and information factor -1.217403092 -1.417070951 Log of technical factor 1.411140358 1.571963524* Log of existing regulation factor 0.137219632 -0.131083337 Log of energy cost 0.381446147 0.669916849** knowledge*split 1.561192111*** regulation*split 0.723553286 split*energy price 32 financial*technical Cutoff parameters omitted Statistics N 98 91 F 1.69E+02 1.32E+02 Variable Model VIII Model IX Log of revenues 2.642495979*** 2.708161900*** Log of energy cost share 0.151742152 0.122248513 Log of financial factor -0.953537008*** -2.322088947*** Log of split factor -1.590755663*** -0.289456429 Log of knowledge and information factor -0.807606837 -0.841238136 Log of technical factor 1.15309456 -0.587985858 Log of existing regulation factor 0.113266895 0.33348211 Log of energy cost 0.088761721 0.389469817 knowledge*split regulation*split split*energy price 1.427506905*** financial*technical 1.055441141** Cutoff parameters omitted Statistics N 98 98 F 55.46015871 2.36E+02 Appendix C: We began by re-estimating the baseline model presented in Table 4, but now missing data regarding firms’ perceived barriers to the adoption were replaced with a zero (Table 1C). The importance of most of the barriers increased by 20% or more relative to the baseline model, but the importance of revenues declined by more than 10%. However, the magnitude of revenues and financial barriers still remained far greater than any of the other barriers estimated. Table 1C. Replacing missing observations in the data with zeros 33 Column1 Column2 Column3 Variable Linear Ologit Total revenues 1.6221*** 2.1925*** Share of energy cost 0.0527 0.1994 Private owned -0.3677*** -0.5879** Financial factor -0.9723*** -1.5305*** Split factor -0.0076 0.0565 Knowledge and information factor -0.6150** -0.8400*** Technical factor 0.6161** 1.0282** Existing rules and regulation factor 0.2425 0.2489 Price of energy 0.2823*** 0.2155* Constant 0.8770*** Cutpoints Cutpoint1 -0.8082* Cutpoint2 1.1169** Cutpoint3 2.4062*** Cutpoint4 3.2745*** Cutpoint5 3.9332*** Cutpoint6 5.2864*** Statistics N 214 214 F 95.0368 15.9024 legend: * p<0.10; ** p<0.05; *** p<0.01 Next, instead of replacing missing-data with zeros, we replaced them with the mean value (Table 2C). The estimated parameters are similar to those obtained when replacing missing-data with zeros. When substituting missing-data with either zero or mean of variable, firm’s revenues, financial barriers, lack of knowledge and low energy prices are the main factors affecting the adoption of energy efficiency measures. Table 2C. Replacing missing observations with mean of variable 34 Column1 Column2 Column3 Variable Linear Ologit Total revenues 1.6088*** 2.1656*** Share of energy cost 0.0396 0.1969 Private owned -0.3308** -0.5381** Financial factor -0.8557*** -1.2845*** Split factor -0.0358 -0.0460 Knowledge and information factor -0.6512*** -0.8892*** Technical factor 0.5396* 0.8963** Existing rules and regulation factor 0.1874 0.1934 Price of energy 0.3082*** 0.2566** Constant 0.9907*** Cutpoints Cutpoint1 -0.8119* Cutpoint2 1.0992* Cutpoint3 2.3917*** Cutpoint4 3.2309*** Cutpoint5 3.8991*** Cutpoint6 5.2299*** Statistics N 213 213 F 90.7612 21.5872 legend: * p<0.10; ** p<0.05; *** p<0.01 35