A comparative and case study analysis of barriers to invest in energy-efficient technologies with an application to the light industry and service sector

M. Koetse

Methodical results

In the methodical part of our research we use Monte-Carlo simulations in order to replicate the actual practice of doing meta-analysis in economic research. We generate primary data inducing a true underlying effect, estimate the primary models, and use the estimated elasticities as effect sizes in our meta-analysis. Subsequently, we estimate several meta-models and analyze how well they capture the true underlying effect, using the bias and mean squared error of the meta-estimators, and the statistical significance of the meta-estimates, as indicators to evaluate estimator performance.
First, we investigate the impact of systematic effect-size variation due to primary-study misspecifications. The standard meta-analytic procedure to deal with this particular problem is to include dummy variables in the meta-model specification. We analyze the performance of this estimator vis-à-vis an estimator that ignores the systematic variation in effect sizes. Second, we analyse the impact of random effect-size heterogeneity by inducing differences in effect-size variance and a varying true underlying effect across studies. Effect-size heterogeneity is accounted for using the effect-size variances as weights in the meta-model specification. We analyze to what extent these estimators mitigate the potentially negative impact of effect-size heterogeneity. Ignoring details and nuances, the results of these two exercises are unambiguous. Standard meta-analytic procedures substantially reduce, but do not fully eradicate, the impact of primary-study misspecifications and effect-size heterogeneity (see Koetse 2006; Koetse et al. 2007a,b).

Investment under uncertainty

The outcomes of empirical studies on the investment-uncertainty relationship allow for only a limited analysis on the magnitude of the relationship. In our meta-analysis we therefore focus primarily on the sign and statistical significance of the outcomes of empirical studies. Although our results do not provide direct evidence on the question whether the relationship between investment and uncertainty is negative or positive, it is clear from exploratory analysis that very few studies actually find positive results that are statistically significant at 5%. This finding is confirmed by our model estimations.
Specifically, the effect-size probabilities show that the probability of finding a positive significant effect size in primary studies is small. Moreover, the marginal effects show that, with a few exceptions, most primary-study characteristics have no or only a very small impact on this probability (Koetse et al. 2007).
Although meta-analysis generally cannot provide decisive insights into the empirical relevance or correctness of certain underlying theories, assumptions and model specifications, it may contribute to the understanding of which factors are actually relevant in explaining the variation in empirical estimates. Our findings suggest that there are several relevant sources of effect-size variation. More specifically, differences in the measurement of investment and uncertainty and empirical differences between primary studies appear to be important factors. In addition, the influence of differences in model specification on the outcome of a study provides evidence on important theoretical issues, and gives clear directions for empirical model building in investment research. The impact of excluding factor prices from investment models on the sign and significance of a primary model estimate is a good example in this respect, but other primary model explanatory variables, such as the financial position of the firm and stock prices, appear to be relevant as well.
Next to the meta-analysis, we performed primary research on the investment uncertainty relationship. In this project we analyze the effects of perceived expectations and uncertainty on firm investment using Dutch firm level data. The basic finding of our analyses is that perceived expectations and uncertainty have a substantial effect on investment spending, and that the specific effect depends on firm size and type of investment. Especially for investment in energy-saving technologies, there is strong evidence for structural differences between small and large firms. Specifically, uncertainty appears to have a larger influence on decision making in small firms than in large firms. However, differences between the two size classes are related to the specific source of uncertainty as well. In small firms, input uncertainty and output uncertainty have a differential impact on both aggregate and energy-saving investments. Moreover, the results suggest that increased uncertainty around wages – a crucial input variable especially in small firms – stimulates factor substitution from labour to capital and shifts attention away from investment in energy-saving technologies. Since expected increases in wages have a similar effect, the results suggest that whenever uncertainty on an important variable such as wages increases, small firms revert to what could be called ‘standard practice’; investing in energy-saving technologies is not one of them (Koetse et al. 2006).

Energy prices and capital-energy substitution

In the context of the meta-analysis on substitution of capital for energy due to energy price changes, an important distinction with respect to the outcomes of empirical studies is between cross-price and Morishima substitution elasticities. The former simply measures the percentage change in demand for capital due to a percentage change in the price of energy. As such, it does not measure the technical substitution potential, because it incorporates income effects. Substitution or Morishima elasticities exclude this income effect and exactly measure the curvature of the production isoquant. Which measure is to be preferred depends primarily on the underlying research or policy question.
Investigations on energy-policy plans may want to focus on actual changes in factor demand, while academic studies on, for instance, technical substitution potential in different sectors of the economy are likely to be more interested in technical substitution measures. We therefore perform an analysis on sign and magnitude of both elasticity measures. In the meta-analysis model we control for differences between the empirical studies in order to uncover the relevant sources of systematic effect-size variation. In addition, we estimate separate coefficients for cross-price and substitution elasticities.
Our findings suggest that omitted variable bias in primary studies is a relevant source of variation. Especially the exclusion of non-neutral technological change parameters affects both the cross-price and substitution elasticity. Furthermore, primary models may produce biased estimates of actual percentage changes in demand for capital when returns to scale and materials are excluded from the model specification. However, pure substitution potential appears to be unaffected by exclusion of these parameters. Therefore, our findings suggest that capital and energy are separable from materials in production functions. The use of cross-section and panel data yields higher estimates of pure substitution potential, probably due to the fact that they more strongly reflect long-run changes in factor demand. Finally, judged by the statistical significance of the effect, the year of evaluation in the primary study does not contribute to explaining variation in the sample of substitution elasticities.
This result suggests that the increasing availability and performance of energy-saving technologies over time has not led to an increase in substitution potential. A potential explanation for this finding is, however, that the bulk of the empirical studies use data from the 1970s and 1980s. The number of effect sizes obtained from empirical studies that use data from the 1990s is therefore too small to warrant strong conclusions on the development of capital-energy substitution potential over more recent time periods.
Taking these insights into account, our results suggest that substitution potential between capital and energy is substantial. Under certain assumptions the elasticity of substitution is equal to .64 in the short run, and increases to .89 and 1.21 in the medium and long run, respectively. However, despite the fact that technical opportunities to substitute capital for energy are considerable, in the short run and medium run they do not outweigh the negative income effect brought about by energy-price increases. The cross-price elasticities are positive but small and not statistically different from zero. In the long run, however, the substitution effect dominates the income effect. Our findings therefore suggest that, although technical substitution potential is large, actual changes in demand for energysaving capital due to energy-price increases take time (Koetse et al. 2007).

Energy productivity convergence across countries and sectors

With respect to the research on energy-efficiency improvements, extensive work has been done on exploring trends and driving forces of energy-efficiency improvements across countries and across sectors. In various papers we analyzed trends in energy productivity in comparison with labor productivity development for 14 OECD countries and 14 sectors over the period 1970-1997. We have tried to determine the most important factors driving these developments, including structural changes, technology driven energy-efficiency improvements, energy prices and wages, economies of scale, investment ratio, openness and specialization. Furthermore, we have analyzed whether there exists a bias towards labor- or energy productivity growth and how this trend evolves over time in various sectors (Mulder and Miketa 2005; Mulder and De Groot 2007).
In addition we have dealt with the question whether cross-country differences in energy- and labor productivity are increasing or decreasing over time. We found that the development of the crosscountry
variation in productivity performance differs across sectors as well as across different levels of aggregation. Both patterns of convergence as well as divergence are found and cross-country variation of productivity levels is typically larger for energy than for labor. Moreover, our analysis provided support for the hypothesis that in most sectors lagging countries tend to catch up with technological leaders, in particular in terms of energy productivity.
In collaboration with A. Miketa we extended this type of analysis for energy productivity to countries outside the OECD, including 56 developed and developing countries, in 10 manufacturing sectors, for the period 1971 to 1995. We found that, except for the non-ferrous metals sector, cross-country differences in absolute energy-productivity levels tend to decline, particularly in the less energyintensive industries. Testing for the catch-up hypothesis confirms that in all manufacturing sectors energy-productivity growth is, in general, relatively high in countries that initially lag behind in terms of energy-productivity levels. At the same time, cross-country differences in energy-productivity performance seem to be persistent; convergence is found to be local rather than global, with countries converging to different steady states and several failing to catch up (Mulder and Miketa 2005).