GCB Bioenergy (2015) 7, 468–478, doi: 10.1111/gcbb.12163
Explaining the reductions in Brazilian sugarcane ethanol production costs: importance of technological change ~ Z 2 and B I N G X U 1 X I A O G U A N G C H E N 1 , H E C T O R M . N U NE 1 Research Institute of Economics and Management, Southwestern University of Finance and Economics, Chengdu 610074, China, 2 Department of Economics, Centro de Investigacion y Docencia Economicas (CIDE), Circuito Tecnopolo Norte S/N, Aguascalientes, AGS 20313, Mexico
Abstract Over the period 1975–2010, unit production costs of sugarcane ethanol in Brazil declined by 67%, while the perunit processing costs decreased by more than 70%. This article examines the role of various factors that lead to these cost reductions, including learning-by-doing (LBD), economies of scale, rising input prices, market competitiveness, and exogenous technological changes. Using the aggregate industry-level data, we show that the traditional experience curve approach will lead to biased estimate of the learning effect when economies of scale, rising input prices, market competitiveness, and exogenous technological changes are excluded as explanatory variables in explaining these cost reductions. With the inclusion of these variables and LBD, we find that the reductions in production/processing costs of Brazilian sugarcane ethanol were primarily driven by autonomous technological changes and unrelated to LBD. Economies of scale, market competitiveness, and rising input prices had insignificant impacts on the reductions in production/processing costs of sugarcane ethanol over the sample period. Keywords: Brazil, learning-by-doing, processing costs, production costs, sugarcane ethanol, technology change
Received 16 September 2013; accepted 16 November 2013
Introduction Biofuels have been supported by governments around the world to enhance energy security, mitigate climate change, and stimulate rural economic development. In Brazil, the primary biofuel supporting policy was the National Alcohol Program (Pr o-Alcool) launched in 1975 that promoted the production of ethanol derived from sugarcane. The program provided several types of supporting policies, including a guaranteed purchase of ethanol by a state-owned oil company, low-interest rate loans to ethanol producers, subsidies for both ethanol and automotive industries, and a mandatory blending of ethanol with gasoline. In the 1990s, the Brazilian government reduced ethanol subsidies and started the industry’s deregulation, but the blending mandate was kept. These supporting policies, together with the commercial success of flex-fuel vehicles (FFV) in the last decade, have significantly contributed to the rapid development of Brazilian sugarcane ethanol industry. Sugarcane ethanol production in Brazil increased more than 47-fold from 0.58 billion liters in 1975 to Seniority of authorship is shared equally.
Correspondence: Bing Xu, Research Institute of Economics and Management, Southwestern University of Finance and Economics, China, tel. +86 28 87099204, fax +86 28 87356958, e-mail:
[email protected]
468
28.0 billion liters in 2010 with an average annual growth rate of 13%. Brazil also became the largest ethanol-producing country in the world before the United States surpassed it in 2007 (Empresa de Pesquisa Energetica, 2011; Renewable Fuels Association, 2012). Over the same period, unit production costs of sugarcane ethanol (including both feedstock costs and processing costs) declined by 67% from Brazil real (R)$2.2 per liter in 1975 to R$0.7 per liter in 2010 in 2005 prices, while the per-unit industrial processing costs decreased by more than 70% from R$1.0 per liter in 1975 to R$0.3 per liter in 2010 (PECEGE, 2011; Van den Wall Bake et al., 2009; CEPEA/ESALQ/USP, 2012). Because of these large cost reductions, sugarcane ethanol is considered as the only economically viable biofuel in the world fuel market (OECD, 2006; Hettinga et al., 2009). Moreover, as sugarcane ethanol has a considerably lower emissions intensity relative to gasoline and other first-generation biofuels (such as food starch-based ethanol and biodiesels derived from vegetable oils), Brazil is considered to have the most sustainable biofuel economy in the world (Goldemberg, 2007). The purpose of this study is to discover the engine that leads to this success by examining the factors that could have contributed to the reductions in production costs of sugarcane ethanol. Several studies in the literature have attributed unit cost reductions of many products and technologies to © 2013 John Wiley & Sons Ltd
E X P L A I N I N G S U G A R C A N E E T H A N O L C O S T R E D U C T I O N S 469 their respective cumulative production experience, referred to as learning-by-doing (LBD). Empirical evidence of LBD was first identified in the airframe industry by Wright (1936) and formalized by Arrow (1962). The relationship between unit production cost and cumulative production experience is represented by Y = axb, where Y is the unit cost of production; x is the cumulative experience which is typically represented by cumulative installed capacity or cumulative production of a product (such as megawatts of electricity generating capacity, megawatt hours produced, and million liters of biofuel production); a is the initial production cost of the first unit; and b is a parametric constant capturing the rate at which cost reductions occur. The learning rate (LR), defined as 2b, is the rate at which per-unit cost of a technology is expected to decline with every doubling of cumulative production. Various learning mechanisms may be operating simultaneously, such as learning-by-searching, learning-by-using, learning-by-interacting, changes in production design and standardization, and spillovers from other activities (see a recent review in Yeh & Rubin, 2012). As these factors often overlap and are difficult to separate, the cumulative production of a technology is often used and serves only as a proxy for a combination of the factors that contribute to cost reductions. Numerous empirical studies have estimated the LR for renewable energy technologies, such as solar, wind, nuclear, and biofuels. These studies show that LR ranges from 0.15 for electricity generated from biomass to 0.35 for solar photovoltaic (PV) (see review studies by International Energy Agency, 2000; McDonald & Schrattenholzer, 2002). In the context of biofuels, Hettinga et al. (2009) examined the relationship between unit production costs of US corn ethanol and cumulative corn ethanol production, and estimated a LR of 45% over the timeframe 1980–2005. Using the same approach, Goldemberg et al. (2004) estimated a LR of 29% for total production costs of sugarcane ethanol over the period 1985–2002. Using a longer time period of 1975–2004, Van den Wall Bake et al. (2009) found a LR of 20% for total production costs of sugarcane ethanol. However, in addition to cumulative production experience, other factors, such as economies of scale, rising input prices, market competitiveness, and technological change induced by research and development (R&D), may also affect production costs of a product or technology. For instance, Isoard & Soria (2001) found that returns to scale are important sources of productivity growth in the solar PV and wind energy industry. Rising input prices are found to have led to energy-efficient innovation and adoption of more energy-efficient technologies in several © 2013 John Wiley & Sons Ltd, GCB Bioenergy, 7, 468–478
industries, such as air conditioning and water heating (Newell et al., 1999) and coal liquefaction and solar energy (Popp, 2002). Studies also suggest that the economic success of the machine tool industry in various Asian industrialized countries/regions can be partly attributed to international competition (Fransman, 1986; Fagerberg, 1988). Moreover, Papineau (2006) and Nemet (2006) found that autonomous technological changes and knowledge spillovers are key drivers reducing unit production costs of solar PV and wind energy, while LBD only has a weak effect. With the inclusion of economies of scale, rising input prices, market competitiveness, an autonomous technological change, and cumulative ethanol production, Chen & Khanna (2012) showed that the LR of US corn ethanol estimated in Hettinga et al. (2009) was biased. The factors mentioned above could also have contributed to the reductions in production costs of sugarcane ethanol. As the number and installed production capacity of sugarcane ethanol mills increased, unit production costs may have moved along the U-shaped average cost curve to the right due to economies of scale, while rising input prices (such as fuels, labor, capital, and sugarcane that are intensively used in the production of sugarcane ethanol) are likely to induce factor-saving innovation. Ethanol production in the rest of the world (ROW) has increased significantly by more than 70-fold from 0.3 billion liters in 1978 to 55 billion liters in 2010 (Earth Policy Institute, 2012). Competition for international ethanol markets could have led Brazil to adopt advanced production technology to increase its international competitiveness and maintain its market share. It is also possible that the reductions in unit production/ processing costs of sugarcane ethanol are induced by R&D-related technological changes only and unrelated to economies of scale, rising input prices, market competitiveness, and LBD. In this study, we quantify the effect of various factors mentioned above on the reductions in production costs of sugarcane ethanol. We first develop a stylized costminimization model to illustrate the factors that may affect average production costs of a product and build our hypothesis. We then perform econometric analysis to examine the signs and statistical significance of these factors in affecting unit processing and production costs of sugarcane ethanol, respectively, over the period 1975–2010. Moreover, we use alternative model specifications and data to examine the robustness of our findings. Our empirical results show that in contrast to Goldemberg et al. (2004) and Van den Wall Bake et al. (2009), exogenous technological change is the key driver, reducing the production and processing costs of sugarcane ethanol, while there is no evidence indicating that rising input costs, LBD, and economics of scale are
470 X . C H E N et al. important factors affecting sugarcane ethanol production costs over the sample period. This article contributes to the existing literature by showing that the estimation of LR is sensitive to the inclusion of additional explanatory variables. It also indicates that the economic success in Brazil’s sugarcane ethanol industry could be driven by exogenous technological development and unrelated to LBD. This article is organized as follows: First, we provide a brief introduction about the sugarcane ethanol industry in Brazil. We then describe the methodology and empirical estimation strategy, followed by the description of data sources. We present the main results of the analysis and discuss our empirical findings in the remaining of the article.
Background: Brazilian sugarcane agroindustry Sugarcane production in Brazil dates back to 1532. It started in the Northeast and then moved to the Sao Paulo area. Currently, Central and Eastern Sao Paulo are two largest sugarcane producing regions in Brazil, together accounting for more than 60% of the nation’s sugarcane production. They also have comparative advantage in producing sugarcane relative to other regions in Brazil. Average sugarcane yield in Sao Paulo was 85 metric tons (MT) per hectare in 2010, while the national average was 79 MT per hectare (IBGE, 2011). Because of the cost advantage, most sugarcane ethanol facilities in Brazil are also located in the Sao Paulo region, producing over 60% of the nation’s ethanol (UNICA, 2012). Sugarcane ethanol industry was considered to be very labor-intensive, but currently more capital- and technology-intensive. It is primarily due to the changes in the labor and environmental regulations (e.g., Sao Paulo state law 11.241 of 2002 and Protocolo Agroambiental de S~ao Paulo of 2007). Initiated in Sao Paulo in 1991, these regulations prohibit among others burning sugarcane leaves and trash before harvesting, which in turn requires implementation of mechanical harvesting (Nyko et al., 2013). Currently, mechanical harvesting of sugarcane is widely adopted in the Southeast region, accounting for about 60% of sugarcane farms, while manual harvesting (which burns dry sugarcane leaves and trash) is mainly practiced in the Northeast region and accounts for another 40% of sugarcane farms. Although these regulations led to an increase in the production efficiency of sugarcane harvesting, they also increased land prices, mainly in the state of Sao Paulo, and the production costs of sugarcane, as pointed out by Gasques et al. (2008). Since May 2000, most sugarcane producers have agreed to sell their production based on the prices set
by COSECANA (Conselho dos Produtores de Cana-de car e Alcool), car, Acßu which calculates retailed Acßu sugarcane prices using the market prices of sugar and ethanol, and the Total Recoverable Sugar (TRS) content of the cane. This agreement aimed to reduce the uncertainty in sugarcane prices after the complete industry’s deregulation in 1999. Therefore, sugarcane prices, mainly in the state of Sao Paulo, depend directly on the TRS prices reported by COSECANA. Table 1 shows the average production costs of sugarcane ethanol in Sao Paulo. As can be seen, the primary cost component of sugarcane ethanol production is feedstock cost, accounting for 63% of the total production costs in 2010. Labor and capital are another two major cost components, which represent 9% and 8%, respectively, of the total production costs in 2010. On average, each MT of sugarcane produces about 0.13 MT of sugar or 85 liter of hydrous ethanol (PECEGE, 2009). Hydrous ethanol can be dehydrated to produce anhydrous ethanol (99.3% concentration), which can be blended with gasoline as transportation fuel. The rapid growth of Brazil’s sugarcane ethanol industry can be largely attributed to government supporting policy initiatives and technological changes in domestic automobile industry. The Brazilian government launched the National Alcohol Program (Pr o-Alcool) in 1975, aiming to substitute part of the automobile fossil fuel with sugarcane ethanol. Initially, anhydrous ethanol dominated the industry to meet the gasoline blending mandate. After the introduction of the first ethanol-dedicated light-duty vehicle (EDV) in 1979, which runs under 100% hydrous ethanol (E100), along with the low sugar prices in the following years, sugarcane ethanol production increased from 3.7 billion liters in 1980 to 11.5 billion liters in 1990 with hydrous
Table 1 Costs of hydrous sugarcane ethanol production in Sao Paulo in 2010* R$/1000 l Raw material (Sugarcane at the mill) Labor Management Chemicals, electrodes, lubricants, and electricity Maintenance (including fuel) Depreciation Returns to Capital Total Cost
Share (%)
501.84
62.5
72.54 55.89 18.58
9.0 7.0 2.3
53.62
6.7
36.05 64.93 803.45
4.5 8.1 100.0
*All costs are converted to 2005 prices using a domestic price index in Brazil. © 2013 John Wiley & Sons Ltd, GCB Bioenergy, 7, 468–478
E X P L A I N I N G S U G A R C A N E E T H A N O L C O S T R E D U C T I O N S 471 ethanol comprising of more than 93% of the total ethanol production. In the 1980s, EDVs accounted for about 80% of the total vehicle sales. However, the sale of EDVs declined drastically from 50% in 1990 to a negligible 0.1% in 1998 (ANFAVEA, 2011) because of the shortage in ethanol supply caused by the removal of the supporting policies and the recovery of sugar prices. In 2003, the Brazilian automotive industry introduced FFVs to the market. FFVs run on any proportion of blended gasohol and E100, and represented more than 70% of the total light-duty vehicle sales during the period 2003–2010 (ANFAVEA, 2011). Together with the high gasoline prices over this period, the development of FFVs led Brazil to resume its ethanol production, reaching historical highs (in average 27 billion liters per year during 2008–2010) and representing 17% of total transportation fuels consumed in Brazil in 2010 (Empresa de Pesquisa Energetica, 2011), with hydrous ethanol accounting for more than 60% of total ethanol production.
Materials and methods We use a cost-minimization model in which one final good (e.g., sugarcane ethanol) is produced with N inputs to illustrate the factors that affect average production costs of the good and to build our hypothesis and empirical estimation strategy.
Conceptual model Consider a representative firm that uses input xit with i e {1, 2, ….N} to produce a final good qt at time t = {t1, t2,….T}. Assuming a Cobb-Douglas production technology, the firm’s N Q production function can be expressed as qt ¼ AðQt ; tÞ xaiti i¼1
with ai > 0, where A(Qt, t) is the total factor productivity (TFP) at time t and an increasing function of cumulative production (Qt) and exogenous technological progress (represented by a N P time trend). Let r ¼ ai . Thus, the production function exhibi¼1
its constant, increasing, or decreasing returns to scale as r = 1, r > 1 or r < 1, respectively. In a competitive market, the firm chooses the optimal combination of xit to produce a given quantity of output qt, while minimizing total production costs. For the ease of illustration, we do not discount total production costs in the objective function. The firm’s problem can be formally stated as follows: Min xit
S:t
X
xit xit
i;t
Að:Þ
N Y
xait1 qt
xit ai xjt ¼ xjt aj xit
where i 6¼ j
ð1Þ
Total costs of production (TCt) can be derived as: TCt ¼
2 P 31=r N aj 6 7 1=r Y 1=r Y j6 ¼ i N N 6 N X 7 qt a =r a =r 6ai 7 ¼ qt : xiti : : xiti :w 6Q 7 N Að:Þ Að:Þ aj 5 i¼1 i¼1 4 i¼1 aj j6¼i
ð2Þ 2P 31=r N aj N 6 j6¼i 7 P 6ai 7 where w ¼ is a constant term. Average costs of 6Q 7 4 N aj 5 i¼1
j6¼i
aj
production (ACt) can be obtained by dividing qt on both sides of (2) as shown in Eqn (3). 1=r1
ACt ¼
qt
1=r
Að:Þ
:
N Y
a =r
xiti :w
ð3Þ
i¼1
Taking the natural logarithm on each side of (3) yields: log ACt ¼
N X 1 1 ai log xit þ log w ð4Þ 1 log qt log Að:Þ þ r r r i¼1
As shown in Eqn (4), average production costs are affected by four components, namely production level (qt), TFP (A(.)), input prices (xit), and a constant term Ψ. ACt is expected to decline with increasing returns to scale (r > 1) and an increase in A(.) that can be achieved by either the accumulation of production experience or autonomous technological change. Moreover, with ai > 0, an increase in input prices would increase average production costs if A(.) remains unchanged. However, increases in input prices could induce the firm to adopt factor-saving technology and/or improve workers’ skills, which would increase A(.) (Newell et al., 1999; Popp, 2002). Therefore, the net impact of the changes in input prices on unit production costs is theoretically ambiguous and requires empirical examination. The experience curve approach in the literature links the changes in average production costs with accumulated production experience (represented by cumulative production), and their relationship is expressed by the following formula: ACt ¼ C0 Qbt PIt
ð5Þ
PR ¼ 2b
ð6Þ
where PIt is price index, such as GDP deflator or consumer price index, used to adjust nominal average costs; b is the experience index; C0 is a constant; and PR is progress ratio, denoting the rate at which unit production costs decline for each doubling of cumulative production. LR is expressed as 1PR. Taking the natural logarithm of (5) yields: log ACt ¼ log C0 þ b log Qt þ log PIt
ð7Þ
i¼1
where xit denotes input prices. The first-order optimality conditions lead to: © 2013 John Wiley & Sons Ltd, GCB Bioenergy, 7, 468–478
Comparing Eqns (4) and (7), we can see that to make (7) hold, one needs to make the following assumptions: (1) constant returns to scale of production (r = 1); (2) cumulative
472 X . C H E N et al. production is the only factor driving the cost reduction 1 (Qbt ¼ Að:Þ r ); and (3) PIt can capture the changes in input prices. However, if these assumptions do not hold and qt, xit, and the time trend are correlated with Qt, using (7) to explain the reductions in unit costs of production will lead to a biased estimate of b.
Empirical estimation strategy In the empirical analysis, we examine factors that could have contributed to the reductions in both unit processing costs and total production costs of hydrous sugarcane ethanol, respectively. As energy, labor, capital, and feedstock costs together account for more than 85% of the total production costs of sugarcane ethanol (see Table 1), we use their market prices to represent input prices. As international competition may also contribute to an increase in TFP (Fransman, 1986; Fagerberg, 1988), we decompose the A(.) into three components: LBD-induced technological changes, exogenous technological changes (represented by a time trend), and market competition (denoted by cumulative ROW ethanol production). We use cumulative ROW ethanol production to represent market competition mainly because about 20% of sugarcane ethanol produced in Brazil is exported to the international markets. Major importers include the United States, Europe, and Caribbean Basin Initiative (CBI) countries. Therefore, the growth in ethanol production in other countries/regions may affect the demand for sugarcane ethanol in the international markets, and thus the competitiveness of sugarcane ethanol. In the empirical analysis, we use Eqn (8) to examine signs and statistical significance of various factors in explaining the reductions in sugarcane ethanol production costs. log ACt ¼ b0 þ b1 log qt þ b2 log Qt þ b3 log Et þ þ b5 t þ et
N X
b4i log xit
i¼1
ð8Þ where subscript t denotes time; ACt is the average production cost of sugarcane ethanol after the adjustment of price index; qt is annual sugarcane ethanol production; Qt is cumulative sugarcane ethanol production; Et is cumulative ROW ethanol production; xit is price of input i (sugarcane prices will be excluded when examining the reductions in industrial processing costs of sugarcane ethanol); variable t is time effect capturing exogenous technological changes due to R&D; and et is the error term. We use ordinary least squares (OLS) method to estimate Eqn (8) under the null hypothesis such that independent variables are exogenous and the error term et is independent and identically distributed. However, our empirical estimation has two major problems that could lead to biased estimates of coefficients. First, autocorrelation in the error term could occur as a result of the use of time series data. Second, the lack of data on other input prices may lead to omitted-variable and endogeneity biases. Besides the explanatory variables included in Eqn (8), other variables could have also affected the production costs of sugarcane ethanol, including the prices of chemical,
enzymes, electrodes, maintenance equipment, and bagasse (the residual cane waste that can be used to produce heat and power). If these variables affected cumulative production, then the assumption E [Qt, et] = 0 would be violated (see KahouliBrahmi, 2008). In addition, like other existing studies that examine technological learning in renewable energy sectors, particularly solar PV, wind, and ethanol (see Isoard & Soria, 2001; McDonald & Schrattenholzer, 2002; Soderholm & Sundqvist, 2007; Kahouli-Brahmi, 2008; Van den Wall Bake et al., 2009), our analysis also uses a small aggregate industry-level dataset. Given the data limitation, our main purpose here is to examine the robustness of the estimated LR to the inclusion of the other factors introduced in Eqn (8) and to compare our findings with those obtained by Goldemberg et al. (2004) and Van den Wall Bake et al. (2009). We conduct a number of tests to examine the appropriateness of our estimation strategy, including Durbin–Watson (DW) and Breusch–Godfrey Lagrange Multiplier (LM) statistics to test for the presence of first-order autocorrelation, and the Hausman test for endogeneity of the cumulative sugarcane ethanol production. When conducting the Hausman test, we use 1-year lag in cumulative sugarcane ethanol production and annual sugarcane ethanol production as instrumental variables for cumulative sugarcane ethanol production.
Data collection Data collected for the econometric analysis range from 1975 to 2010. The dataset includes unit total production and processing costs of sugarcane ethanol, annual sugarcane ethanol production, sugarcane prices, market prices of inputs (including energy, labor, and capital) used for ethanol production in Brazil, and world ethanol production. All prices and costs are converted to 2005 prices using a domestic price index (i.e. Indice Geral de Precßos - Disponibilidade Interna, IGP-DI) reported by Instituto Brasileiro de Economia da Fundacß~ ao Getulio Vargas (2012). With the lack of facility-level data and knowing that Sao Paulo is the primary sugarcane ethanol-producing state in Brazil and supplies more than 60% of the nation’s sugarcane ethanol, we use average production/processing costs reported for Sao Paulo as a proxy for industry-level production/processing costs of sugarcane ethanol in our analysis. Annual sugarcane ethanol production in Brazil is taken from Empresa de Pesquisa Energetica (2011), while world ethanol production is gathered from Earth Policy Institute (2012). We collect sugarcane prices from Instituto Brasileiro de Economia da Fundacß~ ao Getulio Vargas (2012). Total production and processing costs of sugarcane ethanol come from different sources. Between 1975 and 2004, these costs are obtained from the same sources as reported in Van den Wall Bake et al. (2009). We use hydrated ethanol prices in Sao Paulo reported by CEPEA/ ESALQ/USP (2012) as a proxy to represent total production costs of sugarcane ethanol in 2005 and 2006. Production costs after 2007 and industrial processing costs of sugarcane ethanol over the period 2005–2010 are compiled and organized from reports by PECEGE (2008, 2009, 2010, 2011). Although most sugarcane mills are self-sufficient in power supply by burning bagasse, they use other energy-related
© 2013 John Wiley & Sons Ltd, GCB Bioenergy, 7, 468–478
E X P L A I N I N G S U G A R C A N E E T H A N O L C O S T R E D U C T I O N S 473 inputs, such as lubricants and fuels. As a proxy for energy costs, we calculated Laspeyres Energy Price Index based on energy prices and quantities consumed in Brazil reported by Empresa de Pesquisa Energetica (2011), including diesel, oil, natural gas, industrial electricity, and steam coal. With the lack of wages for sugarcane ethanol industry, we use the minimum legal wage in Brazil as a proxy for labor costs. As a proxy for capital prices, we use pig iron prices reported in the statistical yearbook of the Brazilian metallurgical industry by Ministerio de Minas e Energia (2011). Table 2 presents summary statistics for our variables of interests. We find both annual and cumulative sugarcane ethanol production exhibit a great deal of variation. This is intuitively interesting, given that annual sugarcane ethanol production accounts for the movement along the U-shaped average cost curve, while cumulative ethanol production potentially represents the shifts of the average cost curve. Input prices also vary significantly over the sample period.
Results We now present empirical regression results. We estimate Eqn (8) with three different model specifications to examine the statistical significance and signs of various factors in reducing unit processing costs of sugarcane ethanol. Specifically, model (1a) tests the validity of the experience curve approach with cumulative sugarcane ethanol production as the only explanatory variable. Model (2a) adds annual sugarcane ethanol production, input prices, and a time trend to examine if these factors have affected unit processing costs of sugarcane ethanol over the sample period. In model (3a), we incorporate cumulative ROW ethanol production to test the significance of market competitiveness in reducing sugarcane ethanol processing costs. We also use the same model specifications to examine the reductions in total production costs of sugarcane ethanol in models (1b)–(3b) with the inclusion of sugarcane prices as an additional explanatory variable. Standard errors of the estimates of coefficients are reported in parentheses. The R2, adjusted R2, DW test, and P-values for Breusch–Godfrey
LM, and the Hausman test statistics are also reported. Moreover, we conducted Augmented Dickey-Fuller (ADF) tests for both dependent variables to examine whether they exhibit unit roots. The ADF test statistics are 4.255 with a P-value of 0.0037 for per-unit processing costs of sugarcane ethanol and 2.169 with a P-value of 0.031 for unit production costs of sugarcane ethanol, respectively. These test results show no evidence of the presence of unit roots for both dependent variables.
Results on processing costs of sugarcane ethanol Results of the regression analysis for models (1a)–(3a) are presented in Table 3. With cumulative sugarcane ethanol production as the only explanatory variable in model (1a), we find its coefficient is 0.255, and statistically significant at the 1% level, implying an LR of 16%, which is slightly smaller than that (20%) estimated by Van den Wall Bake et al. (2009). That can be attributed to the difference in the data used between the two studies. However, the DW test statistic of 0.584 and the P-value of Breusch–Godfrey LM of 0.001 indicate that the error terms in this model specification are positively correlated (the 5% critical values for DW test with N = 36 and K = 2 are dl=1.41 and du = 1.52; thus d < dl). Moreover on the basis of the Wu–Hausman test statistic, we can reject the null hypothesis of no endogeneity at the P < 0.10 level, which indicates that omitted variables are correlated with cumulative sugarcane ethanol production. Therefore, the evidence suggests that the model (1a) is misspecified and the coefficient estimated in the model (1a) is biased. In contrast, in regressions (2a)–(3a) with the inclusion of additional explanatory variables, the DW and Breusch–Godfrey LM test statistics show no evidence of the presence of serial correlation and the Wu–Hausman test statistics also indicate that cumulative sugarcane ethanol production is not an endogenous variable.
Table 2 Summary statistics* Variable
Mean
SD
Minimum
Maximum
Unit processing cost of sugarcane ethanol (R$ per liter) Unit total production cost of sugarcane ethanol (R$ per liter) Cumulative sugarcane ethanol production (Billion liters) Cumulative ROW ethanol production (Billion liters) Annual sugarcane ethanol production (Billion liters) Sugarcane price (R$ per MT) Energy price index Wage (R$ per month) Pig iron (R$ per MT)
0.61 1.36 156.6 64.3 11.8 45.0 70.6 339.3 587.6
0.27 0.61 128.6 78.0 6.8 18.5 44.1 111.8 194.1
0.23 0.64 1.2 0.02 0.6 23.1 24.6 166.0 356.0
1.07 2.33 427.2 319.4 28.0 86.2 199.2 541.1 1086.4
*All prices and costs are converted to 2005 prices using a domestic price index in Brazil. © 2013 John Wiley & Sons Ltd, GCB Bioenergy, 7, 468–478
474 X . C H E N et al. Table 3 Regression results (dependent variable: Log unit processing costs of sugarcane ethanol) Model
(1a)
Log Cumulative sugarcane ethanol production Log Annual sugarcane ethanol production Log Energy price index Log Wage
0.255 (0.028)***
R-square Adjusted R-square DW test statistics Breusch– Godfrey LM† Wu–Hausman F test†
(3a)
0.344 (0.212)
0.059 (0.339)
0.293 (0.216)
0.366 (0.219)
0.273 (0.130)* 0.092 (0.150) 0.166 (0.171) 0.078 (0.015)***
Log Capital price Time trend Log Cumulative ROW ethanol production Constant
(2a)
0.252 (0.130)* 0.085 (0.186) 0.107 (0.178) 0.067 (0.016)*** 0.226 (0.155)
9.174 (0.315)*** 0.714 0.705
3.842 (1.877)** 0.925 0.910
8.123 (3.343)*** 0.929 0.910
0.584
1.962
1.975
0
0.998
0.985
0.088
0.172
0.240
Standard errors in parentheses; *P < 0.1. †P-values of these test statistics.
***P < 0.01,
**P < 0.05,
In regression (2a) with the inclusion of other explanatory variables, such as input prices, annual sugarcane ethanol production, and a time trend, we find that the coefficient on cumulative sugarcane ethanol production is not significant. The coefficient on energy price index is positive and significant at the 10% level, implying that the increase in energy costs raised unit processing costs of sugarcane ethanol. Other input prices and economies of scale are found to have insignificant impacts on unit processing costs of sugarcane ethanol over the sample period. We find that the time trend coefficient is negative and significant at the 1% level, which suggests that the reductions in unit processing costs of sugarcane ethanol were primarily driven by exogenous technological improvements over time and unrelated to LBD. Here, exogenous technological improvements may include the adoption of advanced
production technology for sugarcane and ethanol as well as the major progress made in auto industries in Brazil. The latter includes the widespread use of EDVs and FFVs, which led to a significant demand for sugarcane ethanol. Signs and statistical significance of the results obtained in model (2a) remain robust when we include cumulative ROW ethanol production as an additional explanatory variable in regression (3a). As shown in the last column of Table 3, the coefficient of the cumulative ROW ethanol production is insignificant, suggesting that the increase in cumulative ethanol production experienced in the ROW did not have a direct impact on reducing processing costs of sugarcane ethanol. Consistent with the findings in model (2a), the coefficient of the time trend is still negative and statistically significant, while the role of LBD and economies of scale in reducing processing costs of sugarcane ethanol is insignificant. Rising energy prices have increased processing costs of sugarcane ethanol, while there is little evidence suggesting that the increases in labor and capital costs have played significant roles in affecting processing costs of sugarcane ethanol. As compared to model (1a), regressions (2a)–(3a) have higher adjusted-R2, indicating that these two models explain a larger portion of the variability in processing costs of sugarcane ethanol over the sample period. We further examine the effect of the deregulation on sugar and ethanol industries in 1999, and the changes in the labor and environmental regulations in 1991 by including dummy variables in model (3a). We find that the effects of these policy changes were not statistically significant; this is possibly because their effects may have been captured by the time trend variable or other variables included in model (3a). The signs and statistical significance of other explanatory variables in this model specification are the same as those obtained in model (3a), which shows the robustness of our results.
Results on total production costs of sugarcane ethanol Table 4 shows regression results on the reductions in total production costs of sugarcane ethanol. We find the coefficient on cumulative sugarcane ethanol production is still negative and statistically different from 0 in model (1b). Again, the DW test statistic of 0.383 and the P-value of Breusch–Godfrey LM indicate the presence of serial correlation of the error terms in this model specification, suggesting that the model (1b) is misspecified and the coefficient estimated is biased. We then include a time trend and other explanatory variables in models (2b) and (3b), respectively, to examine the statistical significance of these variables in reducing total production costs of sugarcane ethanol. © 2013 John Wiley & Sons Ltd, GCB Bioenergy, 7, 468–478
E X P L A I N I N G S U G A R C A N E E T H A N O L C O S T R E D U C T I O N S 475 Table 4 Regression results (dependent variable: Log unit total production costs of sugarcane ethanol) Model
(1b)
(2b)
(3b)
Log Cumulative sugarcane ethanol production Log Annual sugarcane ethanol production Log Sugarcane price Log Energy price index Log Wage
0.255 (0.024)***
0.048 (0.163)
0.197 (0.260)
0.14 (0.168)
0.111 (0.177)
Log Capital price Time trend Log Cumulative ROW ethanol production Constant R-square Adjusted R-square DW test statistics Breusch– Godfrey LM† Wu–Hausman F test†
0.247 (0.177) 0.204 (0.100) 0.081 (0.174) 0.081 (0.136) 0.041 (0.014)** 0.08 (0.120)
9.976 (0.269)*** 0.773 0.766
4.855 (1.400)*** 0.957 0.946
6.485 (2.580)***
0.383
1.696
1.702
0
0.380
0.384
0.594
0.532
0.764
2.0
0.956 0.942
Unit processing costs (R$ l–1)
0.252 (0.171) 0.211 (0.097)** 0.01 (0.152) 0.096 (0.127) 0.045 (0.012)***
evidence suggesting that sugarcane prices, wage, and capital costs have affected total production costs of sugarcane ethanol. The role of economies of scale and market competitiveness is still insignificant in both model specifications, as shown in Table 4. We plot fitted values of industrial processing costs and total production costs of sugarcane ethanol obtained from models (1a) and (3a), (1b) and (3b), respectively, against the cumulative sugarcane ethanol production over the period 1975–2010. As shown in Figs 1 and 2, models (3a) and (3b) provide more accurate fitted values than models (1a) and (1b). The differences between fitted values obtained from models (3a) and (3b) and the observed production costs of sugarcane ethanol are typically 10%. Therefore, models (3a) and (3b) are better at explaining the changes in the production costs of sugarcane ethanol over the period 1975–2010.
Observed unit processing costs of sugarcane ethanol Model (1a) Model (3a)
1.5
1.0
0.5
0.0
***P < 0.01,
0
**P < 0.05,
Similar to the findings in models (2a) and (3a), the time trend coefficients are negative and statistically significant in both model specifications, while the coefficients on cumulative sugarcane ethanol production are not statistically significant. Once again, these results suggest that exogenous technological progress has played a key role in reducing total production costs of sugarcane ethanol and the role of LBD is insignificant. Our results are different from those obtained by Goldemberg et al. (2004) and Van den Wall Bake et al. (2009) as they do not control other factors that could have affected production costs of sugarcane ethanol. The positive coefficient on energy price index in model (2b) indicates that rising energy prices could have increased total production costs of sugarcane ethanol, while there is little © 2013 John Wiley & Sons Ltd, GCB Bioenergy, 7, 468–478
100
200
300
400
500
Cumulative production of sugarcane ethanol (billion liters)
Fig. 1 Model fitted values of unit industrial processing costs of sugarcane ethanol (2005 prices).
4.0
Unit procduction costs (R$ l–1)
Standard errors in parentheses; *P < 0.1. †P-values of these test statistics.
Observed unit production costs of sugarcane ethanol Model (1b) Model (3b)
3.0
2.0
1.0
0.0
0
100
200
300
400
500
Cumulative production of sugarcane ethanol (billion liters)
Fig. 2 Model fitted values of unit production costs of sugarcane ethanol (2005 prices).
476 X . C H E N et al. Robustness tests Results presented above are intuitively plausible. However, sugarcane ethanol production in Brazil involves a very complicated supply chain network, especially for large plants. It includes wage negotiations with union workers, procuring sugarcane and other inputs for ethanol production, and competing sugarcane with global sugar markets. Therefore, our coefficient estimates may be sensitive to the chosen input prices. In this section, we examine the stability of our coefficient estimates with regard to input prices. We first examine the sensitivity of our results for the reductions in unit processing costs of sugarcane ethanol. Specifically, in scenario (1) we use wage in year t1 rather than in year t to examine the reductions in processing costs of sugarcane ethanol in year t. In scenario (2), we consider 1-year lag in both wage and capital prices. In scenario (3), we use all input prices in year t1, including wage, capital prices, and energy price index, rather than in year t, as explanatory variables. As shown in Table 5, across these scenarios, statistical significance, magnitudes, and signs of the coefficient estimates of the time trend are almost identical compared to our baseline results, while LBD is still found to be an insignificant factor in affecting processing costs of sugarcane ethanol. In scenarios (2) and (3), the coefficients of cumulative ROW ethanol production are positive and statistically significant, suggesting that the increase in cumulative ROW ethanol production could have had an adverse impact on the reductions in processing costs of
sugarcane ethanol. This market effect could be reflecting that the expansion of ethanol production in the ROW has led to rising costs of procuring key inputs (such as chemical and enzymes) at a point in time. We then examine the robustness of the coefficient estimates for the reductions in total production costs of sugarcane ethanol (results are shown in Table 6). For this, we consider four scenarios. In scenario (1), we consider 1-year lag in wage. Scenario (2) uses 1-year lag in both wage and sugarcane prices, while scenario (3) considers 1-year lag in all input prices. We also consider a scenario (4), where world sugar prices are used as the explanatory variable to represent feedstock costs rather than sugarcane prices. Consistent with the findings in the baseline case, coefficient estimates for the time trend are robust across these scenarios regardless of the use of input prices as explanatory variables. Therefore, these tests confirm that exogenous technological change was the key driver reducing total production costs of sugarcane ethanol in Brazil, while the roles of other factors are found to be insignificant over the sample period.
Discussion It is important to have reliable estimates of different factors in affecting production costs of renewable energy because it affects the assessment of the costs of government policies to promote renewable energy. The experience curve approach has been widely used to explain the reductions in unit production costs of sugarcane ethanol. However, by disregarding the effects of other
Table 5 Robustness tests (dependent variable: Log unit processing costs of sugarcane ethanol) Scenario†
Scenario (1)
Scenario (2)
Scenario (3)
Log Cumulative sugarcane ethanol production Log Annual sugarcane ethanol production Log Energy price index Log Wage Log Capital price Time trend Log Cumulative ROW ethanol production Constant R-square Adjusted R-square DW test statistics Breusch–Godfrey LM‡ Wu–Hausman F test‡
0.108 (0.317)
0.099 (0.271)
0.163 (0.258)
0.331 (0.225)
0.363 (0.207)*
0.429 (0.220)*
0.238 (0.130)* 0.128 (0.169) 0.108 (0.175) 0.065 (0.015)*** 0.235 (0.144)
0.207 (0.125)* 0.136 (0.165) 0.186 (0.149) 0.065 (0.014)*** 0.235 (0.137)*
0.231 (0.131) 0.184 (0.163) 0.230 (0.148) 0.064 (0.013)*** 0.311 (0.135)**
8.524 (2.988)*** 0.930 0.912 1.986 0.981 0.295
8.205 (2.587)*** 0.933 0.915 2.080 0.738 0.230
8.866 (2.515)*** 0.934 0.916 2.167 0.537 0.442
Standard errors in parentheses; ***P < 0.01, **P < 0.05, *P < 0.1. †Scenario (1) considers 1-year lag in wage; Scenario (2) considers 1-year lag in both wage and capital price; and Scenario (3) considers 1-year lag in energy price index, wage and capital price. ‡P-values of these test statistics. © 2013 John Wiley & Sons Ltd, GCB Bioenergy, 7, 468–478
E X P L A I N I N G S U G A R C A N E E T H A N O L C O S T R E D U C T I O N S 477 Table 6 Robustness tests (dependent variable: Log unit total production costs of sugarcane ethanol) Scenario†
Scenario (1)
Scenario (2)
Scenario (3)
Scenario (4)
Log Cumulative sugarcane ethanol production Log Annual sugarcane ethanol production Log Feedstock price Log Energy price index Log Wage Log Capital price Time trend Log Cumulative ROW ethanol production Constant R-square Adjusted R-square DW test statistics Breusch–Godfrey LM‡ Wu–Hausman F test‡
0.281 (0.240) 0.170 (0.179) 0.280 (0.157) 0.187 (0.098)* 0.168 (0.141) 0.080 (0.132) 0.037 (0.012)*** 0.095 (0.113) 7.124 (2.369)*** 0.958 0.945 1.687 0.375 0.956
0.325 (0.249) 0.085 (0.177) 0.173 (0.172) 0.166 (0.105) 0.026 (0.157) 0.072 (0.138) 0.045 (0.012)*** 0.146 (0.113) 8.766 (2.369)*** 0.954 0.940 1.637 0.263 0.623
0.338 (0.201) 0.001 (0.173) 0.153 (0.167) 0.152 (0.104) 0.022 (0.154) 0.196 (0.117) 0.045 (0.010)*** 0.198 (0.106)* 8.658 (1.987)*** 0.957 0.944 1.718 0.400 0.664
0.151 (0.274) 0.051 (0.183) 0.022 (0.072) 0.170 (0.132) 0.065 (0.149) 0.096 (0.142) 0.048 (0.013)*** 0.091 (0.128) 6.623 (2.801)** 0.953 0.938 1.658 0.319 0.851
Standard errors in parentheses; ***P < 0.01, **P < 0.05, *P < 0.1. †Scenario (1) considers 1-year lag in wage; Scenario (2) considers 1-year lag in both wage and sugarcane prices; and Scenario (3) considers 1-year lag in all input prices; and Scenario (4) uses world sugar prices to represent feedstock costs instead of sugarcane prices. ‡P-values of these test statistics.
factors, this approach may lead to biased estimates of LR due to omitted-variable and potential endogeneity issues. This study quantifies the role of various factors that could have played in reducing the production costs of sugarcane ethanol including LBD, economies of scale, rising input prices, market competitiveness, and exogenous technological changes. We present several new findings. First, in contrast to the findings by Goldemberg et al. (2004) and Van den Wall Bake et al. (2009), we find that when economies of scale, rising input prices, market competitiveness, and exogenous technological changes are included as additional explanatory variables in addition to LBD, the coefficient estimate on LBD becomes insignificant. This suggests that the reductions in processing/production costs of sugarcane ethanol could be primarily driven among others by exogenous technological progress and unrelated to LBD. Second, the effects of other input prices on reducing production costs of sugarcane ethanol, such as sugarcane prices, labor costs, and capital prices, are found to be insignificant. Third, rising energy prices, economies of scale, and market competitiveness (measured by the cumulative ethanol production in the ROW) could have played a role in affecting production costs of sugarcane ethanol, but their statistical significances are sensitive to the chosen model specifications and data. The insignificant roles of rising input prices, economies of scale, and market competitiveness, however, could be attributed to the use of aggregate industrylevel data. Due to the data limitation, one should be cautious when using the results presented in this analysis for any policy recommendations. Our main © 2013 John Wiley & Sons Ltd, GCB Bioenergy, 7, 468–478
purpose in this study is to show that in addition to cumulative production experience, it is important to consider other factors when applying the experience curve approach to explain cost reductions of a new technology or product.
Acknowledgement This project is partly supported by National Natural Science Foundation of China (NSFC) grant number 71133007.
References ANFAVEA (2011) Anuario da Ind ustria Automobilıstica Brasileir. Associacß~ao Nacional dos Fabricantes de Veıculos Automotores. National Association of Motor Vehicle Manufacturers, Brazil. Arrow KJ (1962) The economic implications of learning by doing. The Review of Economic Studies, 29, 155–173. CEPEA/ESALQ/USP (2012) Indicadores de Precßos. Available at: http://cepea. esalq.usp.br/indicador/ (accessed 16 February 2013). Chen X, Khanna M (2012) Explaining the reductions in US corn ethanol processing costs: testing competing hypotheses. Energy Policy, 44, 153–159. Earth Policy Institute (2012) Data Center. Available at: http://www.earth-policy.org/ data_center/C23 (accessed 1 February 2012). Empresa de Pesquisa Energetica (2011) Balancßo Energetico Nacional. Fagerberg J (1988) International competitiveness. The Economic Journal, 98, 355–374. Fransman M (1986) International competitiveness, technical change and the state: the machine tool industry in Taiwan and Japan. World Development, 14, 1375–1396. Gasques JG, Bastos ET, Valdes C (2008) Precßos da Terra no Brasil. XLVI Congreso da Sociedade Brasileira de Economia Adminitrac¸ao e Sociologia Rural. Goldemberg J (2007) Ethanol for a sustainable energy future. Science, 315, 808–810. Goldemberg J, Coelho ST, Nastari PM, Lucon O (2004) Ethanol learning curve—the Brazilian experience. Biomass and Bioenergy, 26, 301–304. Hettinga WG, Junginger HM, Dekker SC, Hoogwijk M, McAloon AJ, Hicks KB (2009) Understanding the reductions in US corn ethanol production costs: an experience curve approach. Energy Policy, 37, 190–203. IBGE (2011) Producß~ao Agrıcola Municipal. Available at: http://www.sidra.ibge.gov.br/ bda/agric/default.asp?z=t&o=11&i=P (accessed 3 March 2012).
478 X . C H E N et al. Instituto Brasileiro de Economia da Fundacß~ao Getulio Vargas (2012) Indicadores de Precßos.
Papineau M (2006) An economic perspective on experience curves and dynamic economies in renewable energy technologies. Energy Policy, 34, 422–432. car, Acßu car e Etanol no PECEGE (2008–2011) Custos de Producßa~o de Cana-De-Acßu
International Energy Agency (2000) Experience Curves for Energy Technology Policy. pp. 45–75. Organization for Economic Co-Operation and Development, Paris. Isoard S, Soria A (2001) Technical change dynamics: evidence from the emerging renewable energy technologies. Energy Economics, 23, 619–636. Kahouli-Brahmi S (2008) Technological learning in energy-environment-economy modelling: a survey. Energy Policy, 36, 138–162. McDonald A, Schrattenholzer L (2002) Learning curves and technology assessment.
Brasil: Safra 2007/2008, 2008/2009, 2009/2010, 2010/2011. Popp D (2002) Induced innovation and energy prices. The American Economic Review, 92, 160–180. Renewable Fuels Association (2012) Industry Statistics. Available at: http://www.ethanolrfa.org/pages/statistics (accessed 13 September 2013). Soderholm P, Sundqvist T (2007) Empirical challenges in the use of learning curves
International Journal of Technology Management, 23, 718–745. Ministerio de Minas e Energia (2011) Statistical Yearbook of the Brazilian Metallurgical Industry. Secretaria de Geologia, Mineracß~ao e Transformacß~ao Mineral, Brasilia, Distrito Federal - Brasil. Nemet GF (2006) Beyond the learning curve: factors influencing cost reductions in photovoltaics. Energy Policy, 34, 3218–3232.
for assessing the economic prospects of renewable energy technologies. Renewable Energy, 32, 2559–2578. UNICA (2012) Unica Data. Available at: http://www.unicadata.com.br/ (accessed 13 September 2013). Van den Wall Bake JD, Junginger M, Faaij A, Poot T, Walter A (2009) Explaining the experience curve: cost reductions of Brazilian ethanol from sugarcane. Biomass
Newell RG, Jaffe AB, Stavins RN (1999) The induced innovation hypothesis and energy-saving technological change. Quarterly Journal of Economics, 114, 941–975. Nyko D, Valente MS, Milanez AY, Tanaka AKR, Rodrigues AVP (2013) Evolucß~ao do car. Bioenergia. BNDES Setorial, plantio e da colheita mecanizados da cana-de-acßu 37, 399–442. OECD (2006) Agricultural Market Impacts of Future Growth in the Production of Biofuels.
and Bioenergy, 33, 644–658. Wright TP (1936) Factors affecting the cost of airplanes. Journal of Aeronautical Sciences, 3, 122–128. Yeh S, Rubin ES (2012) A review of uncertainties in technology experience curves. Energy Economics, 34, 762–771.
© 2013 John Wiley & Sons Ltd, GCB Bioenergy, 7, 468–478
