! ! !
!
!
!
!
The dawn of the solar panel?
Revisiting solar prices using a quality-adjusted price index
N.C. Scholze S2172046 Thesis MSc IE&B
Faculty of Economics & Business Supervisor: Dr. T.M. Harchaoui
Abstract
As one of several disruptive technologies lurking in the darkness, solar power has been identified as having the potential of boosting economic growth in the 21st century. However, while various authors have suggested that photovoltaics have experienced a
‘Moore’s-Law-like’ price reduction, there has been little research into actually measuring solar price
reductions compensated for the quality change that has followed from technological progress. Therefore, this paper attempts to arise to the challenge of constructing a quality-adjusted price index for solar panels by using a hedonic pricing model. By holding fixed progress in the constituent technological characteristics of solar panels, hedonic regression estimations offer an accurate method for such analysis. With an average annual growth rate of -20.6% over the time period 2004-2013, findings suggest that taking into account quality-adjusted prices reflects an even brighter future for the economic viability of solar power.
1. Introduction
The aftermath of the 2008 debt crisis is a time of both uncertainty and confusion for economic growth analysts. While a consensus has been reached on the central role of technology in the modern age of economic growth, sceptics represented by Gordon (2014) recently emerged suggesting that with the passing of the 20th century, an era of exceptional growth has ended.
However, doubts remain that new technologies keep unfolding at a staggering rate, with the potential of fundamentally changing current business practices. Following the unparalleled emergence of the microchip and computer processing, digitization has become the modern day mantra, with technologies such as artificial intelligence and network communication taking centre stage. These ‘disruptive technologies’, as McKinsey (2013) labels them, are Schumpeter’s creative destruction at work and offer a way forward to propelling us to 21st century economic growth.
Whatever the outcome of the discussion, economists have historically struggled with accurately predicting what the future might hold. Not only is technological innovation difficult to grasp and measure, it is made even more challenging considering its pace, breadth and potential for spillovers. This daunting task is made increasingly difficult by the fact that official statistics always emerge with a considerable lag relatively to technology. For example, during the early stages of the computer revolution, Robert Solow (1987) remarked that “you see the computer revolution everywhere, except in the productivity data”. Not until the 1990s it was that economic literature picked up on information technology as the crucial lever for the increase in productivity (Jorgenson and Stiroh, 2000). Propelling this revolution was Moore’s Law - a steep decline in the prices of semiconductors. Although the role of information technology at the forefront of technological progress and economic growth is now well understood, it does not remove suspicion that there may be multiple sources driving the global economic growth of tomorrow.
McKinsey Global Institute (2013) has recently identified twelve disruptive technologies that offer the prospect of triggering massive economic transformation in future years. One of these technologies is solar power. To see the scope of its potential, consider the following. Solar power is unique in its ability of empowering people at remote areas, be it on a station in far outer space, a desert in Australia or a rooftop in Germany. In the developing world, severely characterized by a lack of adequate infrastructure, solar power is an enabler, and it may even offer Africa a ticket to economic growth.1 On the other side of the globe, the developed world has realised that replacing depletable energy sources with sustainable power generation methods will be a key challenge to humanity in the 21st century.2 As Thomas Edison put it: “I’d put my money on the sun and solar energy. What a source of power, I hope we don’t have to wait before oil and coal run out to tackle that” (Rogers, 2007).
Fuelling the speculation about the economic viability of solar power as a game changer is the fact that recently, solar power has been on the verge of cracking the price competitiveness of !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
1!While many African people lack basic living necessities, what they are desperately seeking appears to be
connectivity. With 635 million mobile-subscriptions in Sub-Saharan Africa in 20141, use of Internet on mobile
phones is expected to 20-fold by 2020 (The Guardian, 2014. See: http://www.theguardian.com/world/2014/jun/05/internet-use-mobile-phones-africa-predicted-increase-20-fold)
Thus, Internet not only provides a means of communication, it also provides access to the developed world, knowledge and education. Since all these devices require electricity, people at remote off-grid locations are massively turning their attention to solar power. See also Ondraczek (2014) on this. !
2 Global energy demand, especially driven by growth of non-OECD countries, is projected to rise from 524
conventional energy sources (Frankel et al., 2014). At the core of solar power’s imminent arrival is its rapid price decline, a development that has actually been evident for decades and that is reminiscent of that of the information technology revolution. Paul Krugman (2011) argued that “we are (…) on the cusp of energy transformation, driven by the rapidly falling cost of solar power”. Krugman referred to this development as solar energy’s variant of Moore’s Law. Although indeed, by definition, Moore’s Law can only apply to electronic devices that contain transistors, James Hutchby (2013) established that over the course of history prices of photovoltaic technologies have followed a similar trajectory.
Over the last several decades, price declines have been coupled with considerable technological progress in terms of efficiency, power and reliability. From $300 dollars per Watt and 4,5% efficiency in 1960, solar energy has declined to under $1/W with efficiency rates around 16% (Kanellos, 2011). In laboratories, scientists today even achieve efficiency rates above 40% (Naam, 2011). As a result, spurred on by government incentives and stiff competition, global solar power installations have grown at staggering pace in recent years, with global installed solar capacity increasing from 1.3GW in 2000 to 139GW in 2013 (table 2.1, appendix). Thus the PV-sector has turned from a “cottage industry centred in Germany to a $100 billion-dollar business with global reach” (Aanesen et al., 2012).
With the pivotal role of governments in kick-starting the industry being widely acknowledged, several papers have attempted to shed light on the direct drivers behind solar price reductions. Following empirical research by Jarmin (1994), researchers have stressed the importance of economies of scale and the learning curve as driving the progress in the solar industry. Benthem et al. (2008) developed a model of technological advancement of solar technology based on learning-by-doing and environmental externalities such as knowledge spillovers. Others have recognized the effects of dynamics of government subsidies on economies of scale and the learning curve (Shrimali and Baker, 2011). Gregory Nemet (2006) contradicted the importance of the learning curve, but found empirical support for economies of scale, technological progress and the cost of silicon. Yu et al. (2012) also find a strong relation between silicon prices and solar panel prices. In one of the latest contributions in the field, Pillai (2014) specifically points to four factors as the fundamental levers in solar-power cost reduction: (i) reduction in cost of principal raw materials; (ii) the emergence of China as competitive producer; (iii) technological innovation and (iv) increased industrial investments.
Whatever the direct drivers of solar prices, exploring the future potential of solar power as a disruptive technology requires unravelling the price developments of solar power while keeping track of technological progress. Since rapid technological progress induces innovation and quality change, observed price trends may be misleading when unadjusted for that quality change. Despite the efforts of researchers and industry analysts, such price estimation is a particularly challenging task when researching an industry as fragmented, dynamic and heterogeneous as the solar-sector. In this paper, I suggest to tackle that problem by decomposing the technical characteristics of solar panels, which provide a convenient point of entry in tracking recent industry achievements.3
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
3 The costs for installed solar panels constitute two elements: (i) the solar panels, and (ii) the so-called
Using price estimations on basis of cost data, along with information of product technicalities drawn from a variety of sources, the estimation of parameters in hedonic regressions allows for the measurement of quality-adjusted prices by holding fixed the progression of the technological characteristics of the product. Thus, the construction of a price index based on pure price changes follows from the application of a hedonic pricing model. In doing so, I build on work by Berndt et al. (1995) and Jorgenson (2001), who have in similar fashion applied hedonic price estimations to information technology. In light of current solar industry developments, there is a need to push current literature into such uncharted territories. Thus, this paper relates to work by Berndt et al. (1995), Jorgenson (2001), Hutchby (2013) and Pillai (2014).
This paper will proceed according to the following outline. In section II, I will provide an overview of the global solar industry. Due to fast-paced developments and quickly changing trends, mapping the evolution and current state of the global photovoltaic (PV) industry is challenging, but important as background information in understanding solar panel price developments as well as quality changes. Equally challenging is the collection of data, for which there are no easily accessible and well-established data sources (section III). Section IV describes in detail the modelling strategy. Subsequently, section V discusses the results and section IV provides concluding remarks as well as suggesting areas for future research. 2. The Industrial Organization of the Solar Panel
In discussing the industrial organization of the solar panel, a good starting point is to review the astonishing growth trajectory the industry has followed over the last few years. Consequently, I discuss the technology of solar panels in section 2.2 and the industrial set-up of activities related to solar panel production in section 2.3, both of which are relevant in understanding the observed price reductions of solar panels.
2.1 The rise of the solar industry
Since the innovation of the first solar cells for the empowerment of space satellites early last century, the industry has come a long way. Figure 2.1 (appendix) reveals that installed capacity of solar power has grown almost exponentially over the last decade. Table 2.1 summarizes the according data of the growth path represented in figure 2.1; from roughly 1.3GW of solar power in 2000, global solar power reached almost 139GW in 2013. Considering the enormous research and development costs, free markets would have been unable to ignite the solar industry.4 It is for this reason that government policy has had special importance; it funded the first initiatives of research before heavily subsidizing the first private firms that commercialized solar power. On the supply side, long-term loans at low interest rates provided initial support for production facilities, and since then have been coupled with R&D funding, technology transfers, production subsidies and industry research consortia. On the consumption side, tax credits and feed-in tariffs (FiTs) have fulfilled decisive roles (Solangi et a., 2011).
Several authors (e.g. see Haley & Schuler, 2011; Aanesen et al., 2012; Zhang et al., 2014) have described the growth of the solar industry as going through a boom-bust cycle, a point that becomes evident when examining the growth rates of installed capacity in table 2.1 (appendix). Although almost all years feature positive growth rates of well above 30%, there !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
is considerable disparity in those growth rates between years. While this boom-bust growth path has been amplified by volatility in silicon prices, a consensus has emerged that shifting market dynamics have fundamentally changed the industry. The rise of China as a cost-effective manufacturer of solar panels has put pressure on the profit margins of European and U.S. producers. China has even been accused of dumping, using production subsidies to gain a foothold in global solar exports (Zhao et al, 2014; Ming et al., 2014).5 Combined with
regulatory uncertainty, this geographical shift in production has impeded the progression of the industry from a technical point of view, as the relative consolidation of the industry in 2013 has hampered solar panel companies to invest and improve upon their technologies (EPIA, 2014). In years to come this issue should be resolved, as the global cumulative installation is expected to at least double between 2014 and 2018 (EPIA, 2014).
2.2 Technology
With technology being important in driving solar price declines, a short review of solar technology basics is helpful. Solar cells depend on the photovoltaic effect, a natural phenomenon in which silicon, an element found in sand, provides energy via an electric charge when exposed to the photons in sunlight. Thus, silicon is a central input to a solar cell. In order for the crystalline in silicon to be able to conduct that current, the solar cells are purified and consequently covered in an anti-reflective coating. In the next production phase they are connected in series to form a joint circuit, sealed with silicon rubber and covered by a glass panel for protection. Lastly, the cells are interconnected via wiring and subsequently mounted to an aluminium frame, which yields a solar panel (also frequently referred to as solar module) as we recognize it from rooftops (Zipp, 2013).6
At present the dominant PV technology is crystalline silicon, covering 80% market share. Notwithstanding the fact that the technology has a proven track record in reliability, durability and potential for integration into buildings and windows, researchers have also shifted their focus towards thin-film cells and concentrated PV cells that have different characteristics (Lorenz et al., 2008; European Commission, 2014). All three technologies – crystalline silicon, thin-film cells and concentrated PV - generally use some type of silicon, but it is the purity of silicon that affects both efficiency and the production costs of the solar cell, as a higher purity of silicon results both in higher efficiency and a higher production cost.7 Despite higher production costs, there are some obvious advantages to higher efficiency cells compared to cheaper, low efficiency cells: more efficient solar cells require less material, weighs less, are cheaper to export and are required in fewer numbers (Lorenz et al., 2008). Quite simply, a solar panel that is twice as efficient in terms of electricity output compared to another solar panel requires only half the space of that other solar panel. Thus, both silicon (as a crucial input to solar cells) and efficiency (as a key quality characteristic) are essential in analysing solar panel prices.
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
5 China has been involved in a trade case with the US following dumping allegations by the Oregon-based
SolarWorld. In 2012, the US imposed high punitive tariffs of 31% on solar cells from China. In response, Chinese firms bought Taiwanese solar cells to circumvent US tariffs, triggering another antidumping duty petition. Early January 2015 though, the US department ruled that US tariffs potentially could be cut in half as China’s dumping might be lower than initially estimated (SEIA, 2013; Wesoff, 2015).!!
6 See also: Saga (2010) and Solar Panel Info, ‘How are Solar Panels made?’. Website:
http://www.solarpanelinfo.com/solar-panels/how-are-solar-panels-made.php, accessed 30-01-2015.
7!Due to data availability, this study only includes crystalline silicon and thin-film solar cell panels. However, it
In improving the quality of solar panels, technological improvement takes centre stage. Saga (2010) sheds light on the various PV technologies, their historical technological advances and their future prospects. Prime cost reductions have come from process technology improvements such as photolithography. This has not only increased efficiency rates, it has simultaneously allowed economies of scale in areas such as wire-slicing of silicon ingots. In order to achieve cost parity with regular energy sources, the central challenge to the PV industry today is to lower production costs and increase economies of scale on the one hand, while improving cell efficiency via better technology on the other hand. This interdependency is illustrated by the fact that many best-in-class technologies with 25% and higher efficiencies are not suitable yet for the cheap, automated processes that develop current day 15% efficiency solar cells. Saga (2010) estimates that modules should be priced below $1/W for solar power to be cost competitive with conventional energy sources. In recent years that barrier has already been cracked, but cost competitiveness with conventional energy sources is very much dependent on local factors. However, when the newest technologies materialize and market solar panel efficiencies break the 25% barrier, the industry will make a huge leap in becoming globally cost competitive.8
2.3 Industrial Organization
With some of the essentials on solar cell technology in mind, this part of section two turns to the industrial organization of the solar panel, which should provide further comprehension of the price developments of solar panels. Here it is useful to elaborate on the different price quotations that add to the opaqueness of the industry, since regularly solar prices are stated in terms of installed condition. That is, apart from the purchasing costs of the module, installed solar prices also involve costs non-associated with technology and production, but related to services such as financing, installation, maintenance and warranty. These “residual costs” are often labelled BOS: “balance-of-systems costs” (Platzer, 2012). Although there have been significant cost reductions in BOS, this paper will only focus on cost reductions related to the solar panel and its technology. For one thing, BOS prices are very difficult to compare as a result of the fragmented downstream activities (consider figure 2.2, appendix), not to mention the difficulties of data collection. Secondly and more importantly, because balance-of-system cost do hardly relate to technological progress, it is out of scope of this research.
The difference between solar panel prices and BOS costs provides a good foundation though for an understanding of the solar industry’s production value chain, which has been described by several researchers, amongst which Platzer (2012), Haley and Schuler (2011) and Ming et al. (2014).9 As I noted earlier, silicon is a crucial input to the production of solar panels, and thus the upstream chain activities mainly feature the production and purification of silicon, which is an extremely concentrated market. As a result of large set-up costs of silicon plants, the construction of each new plant greatly affects global supply. Combined with volatile demand for silicon, this results in great peaks and falls of silicon prices (figure 2.4, appendix).10 The midstream processes involve all process relevant to solar cell and panel
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
8 De la Tour et al. (2013) estimate a 67% module price reduction between 2011 and 2020, implying a module
price well below 0,60$/W. For a more elaborate understanding of solar cell technology improvements, please review Saga (2010). This section is provided as background knowledge.
9 This is graphically summarized in figure 2.3, appendix.
10 Underlying the direct causes of volatility in the market might also be more structural shocks, a point that Yu et
production, and thus also relate to the processes as touched upon earlier when referring to Saga (2010).
In these midstream activities competition essentially circles around the interaction of price and product quality. While at present there are still many solar panel producers, Pillai & McLaughlin (2013) find that these are very heterogeneous in terms of performance and that there is a considerable dispersion in the mark-up of prices, where larger firms are able to generate economies of scale and can apply larger mark-ups. Furthermore, products are found to differ considerably in terms of quality as measured by technical performance. Consequently, efficient producers with superior quality panels are at the moment driving out unproductive manufacturers. At the downstream end of the value chain such a development is not the case. Here one finds all residual activities as captured by balance-of-system costs, which mainly consist of services. Since these activities are largely influenced by local market conditions such as government regulations and tax incentives, the importance of local market knowledge has prevented companies from dominating this section of the value chain. As a result, downstream processes are greatly fragmented and customers tend to rely on local companies for the purchase, installation and maintenance of solar panels (Platzer, 2012). In summary, before turning to the data collection for the analysis of solar panel price reductions, one should bear in mind the following points for take-away. While governments have fulfilled crucial roles in igniting the industry, a lack of consistent policies has contributed to boom-bust cycles of industry growth, a phenomenon that has been amplified by the volatile price of silicon. As a result, firms have been impeded in investing in research and development and improving upon technologies. Since firms mainly compete on basis of cost and solar panel efficiency, economies of scale are crucial in gaining a foothold in the industry. However, while new and innovative solar panel technologies are available, current-day automated processes require costly adaptations for those technologies to materialize. Meanwhile, price reductions have accelerated owing to the rise of China as a cost-effective solar panel manufacturer.
3. Data
3.1 Data collection
Since there are no readily available datasets in this area, this paper will draw upon the use of a self-composed dataset for its empirical research. Specifically, the non-existence of historical price data per specific solar panel forces me to rely on the average module price per company per year, which I subsequently relate to the average performance of solar panels produced by that company in that respective year. In doing so the construction of this dataset primarily rests on two sources. The first source entails the estimation of yearly average modules prices, which I infer by applying mark-ups on an estimated cost level. To this end, I follow the method used by Pillai (2014) for the estimation of the yearly average cost of solar modules ($/W)11 based on dividing cost of goods sold (COGS) by volume of shipments (measured in MW12). This set of information is available from the financial statements filed by the
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
11 Solar prices are stated in terms of dollars per Watt of peak power. Peak power implies the amount of
electricity (watt) the solar cells can generate per square meter under optimized test conditions. See: Sol Stats:
‘What is solar peak power?’. Website: http://www.solstats.com/blog/solar-energy/what-is-solar-peak-power/,
accessed 05-02-2015
companies included in the dataset. Subsequently, I apply mark-ups from similar-sized solar panel manufacturers that did report yearly average module prices.
In order to have as many observations as possible, data has been collected on sixteen of the largest solar module producers for which data was available. Indirectly, this has also determined the ten year time period (2004-2013).13 Since COGS include all expenses on
material, personnel and other operating processes related to the annual production of goods, for my price estimation to be valid it is essential that all companies included in the dataset are pure solar module manufacturers. Before the early 2000s, the majority of solar modules was produced either by conglomerates (such as Kyocera and Sharp) or petroleum multinationals (such as BP and Shell), neither of which reports data on its solar divisions separately in annual reports. Thus, the companies included in this dataset only feature pure solar companies, and the resulting dataset yields manufacturers from Canada, China, Germany, Norway and the US that differ considerably in terms of revenue, cost, market share and market entry.14 Combined with the fact that in several cases data is lacking for distinct years, this results in a dataset that is an unbalanced panel.
All companies included in the research report their cost of goods sold and annual module shipments (MW) in their annual reports. For the US-based and listed companies, annual reports are available in terms of 10-K Forms that are required for filing to the US Securities and Exchange Commission (SEC), while foreign firms that are listed on US stock exchanges have to file a similar 20-F statement. For the European firms data was available from their respective annual reports, although some of them do not report cost of goods sold but Earnings Before Income and Taxes (EBIT). By subtracting EBIT from annual revenue and summing expenses on personnel and other operating expenses, a similar metric to cost of goods sold is achieved. For those non-US firms that have reported their income statements in local currency, cost of goods sold are transformed into dollar denominations using average annual exchange rates.15 As an imperfect measure of real price, I subsequently estimate mark-ups on cost for those companies that did not report yearly average module prices. For Yingli, I use mark-ups from a similar-sized Chinese manufacturer, Trina Solar. In accordance with findings from Pillai & McLaughlin (2013), Chinese manufacturers enjoy a cost advantage relative to the non-Chinese manufacturers. As a result, I am forced to apply the mark-up of Canadian Solar to the German and Norwegian firms for price estimation.
The second part of the data collection is derived out of the Photon Database and provides the technical characteristics of the solar panels produced by the companies included in this research. The data collected include basic indicators on solar technical performance (module efficiency and power watt peak), as well as reliability (years of product guarantee), longevity !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
13 The firms included are Yingli, Trina, Canadian Solar, First Solar, JA Solar, Jinko Solar, (Kyocera), REC
Solar, Sunpower Corp., LDK Solar, SolarWorld AG, Aleo Solar, Suntech Power, Evergreen Solar, Solar-Fabrik AG and CentroSolar AG. Of the ten largest solar module producers reported in 2.3 in the appendix, Flextronics, Hanwha-SolarOne and Solar Frontier are not represented due to a lack of data available.
14 Market share is calculated by dividing firm MW production in year i by global solar MW production in year i.
Global solar production data is retrieved from Earth Policy Institute. Since this database only covers years up to 2012, the 2013 estimate is retrieved from GTM Research website, one of the prime institutes on which Earth Policy Institute relies for its data. See also: http://www.greentechmedia.com/articles/read/Global-2013-PV-Module-Production-Hits-39.8-GW-Yingli-Leads-in-Production-a
15 Among the Chinese firms these are Jinko Solar and JA Solar, which both report average annual exchange rates
(percentage of power guarantee after 25 years) and size (dimensions and weight). There are two primary reasons to select data from the Photon database. First of all, the German-based Photon is a renowned and respected publisher and conference organizer in the field of solar power.16 Since Photon objectively and accurately measures solar module performances in its outside and inside laboratories, data provided by Photon may be more reliable considering that solar module manufacturers may have a tendency to overstate technical performance. Secondly, the Photon database is the most extensive and detailed database in the industry to include both vintage and current-day solar module specifications.
Before turning to the calculation of the average performance of solar panels per firm per year, it is necessary to discuss a major problem that arises from the Photon database. While there are very many producers and solar panels available, a large share of these panels lacks specifications on vintage (year of start of production) and/or product life cycle (year of end of production. In most instances that a year specification is stated, these primarily concern the year of start of production. Thus, of the 2723 solar panels available in the Photon Database for the sixteen companies included in this research, only 831 panels were found to have full data on the technical specifications required as well as a specification on years of production (be it either starting year of production, ending year of production or both). All panels for which data was either not fully available or panels that did not have any reference at all to their specific time period of production have been discarded from the dataset. Outliers have not been considered.17
Since the calculation of average performance of solar panels per year requires specific reference to both the vintage year and the year of end of production, it is necessary to fill the voids on year specifications that arise in the Photon database. In several instances, a year specification can be deducted on basis of the panel’s respective production series18 for which
data on year specification is available.19 While this solves a piece of the puzzle, the majority of the year specification voids require estimation. In doing so, the following method has been applied. By calculating the product life cycle for those panels for which year data on start and end of production was available, an average lifespan of rounded three years was found for these sixteen panel producers. This product life cycle has subsequently been used to complete the dataset in terms of data entry on years.20
A further complication in the computation of average performance of solar panels arises from the lack of sales data per specific panel per company, which prevents one from giving weights to each solar panel. As a second-best method, geometric means provide a way forward in dealing with this issue. Contrary to arithmetic means, geometric means are better suited to normalize different ranges of data (i.e. in this case the different ranges of the various characteristics of solar panels) and thus should result in a relatively more reliable mean !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
16See also: http://www.photon.info/photon_laboratory_en.photon?ActiveID=1117!
17 Some panel producers sell comparatively very small, lightweight solar panels that have a technical
performance (far) below the average of standard-size solar panels. Therefore, solar panels lighter than 8 kilograms have been treated as outliers and for obvious reasons have not been considered.
18 This is comparable to the choice one has for other products which are produced as a series; e.g. in terms of
engine choice for a Volkswagen Golf or memory storage on an Apple iPhone. Series subsequently relate to production years, such as the Mark 7 for the Golf or the iPhone 6 for Apple.
19 For some panels, verification on years of production is possible using (i) Photon Annual Testing Magazines,
available on its website; (ii) PV Matrix Database, http://www.pv-matrix.de/; (iii) ENF Solar database, which also includes PDF files with technical details provided by the manufacturer. See: http://www.enfsolar.com/pv/panel
20 JA solar is the exception here. Due to limited panel and year availability only year of introduction has been
estimation. In calculating the geometric means of the technical characteristics of a solar panel in year i, all solar panels that are observed in that year are taken into account (regardless of their phase in the product life cycle). This yields one observed solar panel per firm per year that should resemble the average performance of all solar panels that were sold that year. Bearing in mind the fact that this dataset has been self-composed using various different sources of data, I acknowledge that this section of the research inherently is particularly vulnerable and a discussion of its limitations therefore is relevant. While it would have been desirable to have specific data in terms of prices and performance for each specific solar panel, no such database exists for time-series data. Being amongst the first to research this particular area, the method I have followed is, to the best of my knowledge, the best alternative. By discarding all solar panels for which no full data or year specification was available, one of the first limitations is that the dataset may suffer from selection bias. Secondly, the dataset could be considerably improved by giving a weight to each type of solar panel on basis of numbers sold, which will certainly be influenced by the solar panels’ respective phase in the product life cycle. Such a calculation would indeed be a more accurate and reliable method than using geometric averages.
Although most data caveats can be found in the data collection on technical performance, some remarks are in order with regard to the cost and price estimations. Despite the fact that all listed companies that either file 10-K statements, 20-F statements or annual reports will be under intense scrutiny by investors and industry analysts, companies may have misreported on costs or annual shipments. Additionally, given that the industry is particularly heterogeneous, product cycle estimations and mark-up estimations on cost to retrieve prices should be interpreted with care.
3.2 Data summary and descriptive statistics
An overview of the variables, their definitions and their function is provided in table 3.1 (appendix), while the data collected for this study is summarized in table 3.2. Apart from prices and technical features, data has also been collected on several control variables. Since the importance of silicon prices has been often stressed in literature (Yu et al., 2012) I retrieve silicon prices as measured in US$ per metric ton from the US Geological Survey. Additionally, data on global production in MW has been collected from the Earth Policy Institute, since production data can serve as a proxy for the learning curve as researched earlier in literature, while firm production may approximate economies of scale. Furthermore, a dummy variable on Chinese origin is included. Bearing in mind recent allegations on dumping by Chinese manufacturers, this variable should help in giving insight into both the validity of such allegations as well as the global market impact as a result of dumping. Lastly, I have created dummies for each year observed.
Table 3.2: Data summary (2004-2013)
Variable N Mean Std. Dev. Min Max
firm 112 8.616071 4.706227 1 16 year 112 2009.17 2.441711 2004 2013 cost 112 2.526875 1.641581 0.53 9.05 price 43 1.881163 1.227163 0.5 4.23 power 112 194.3175 50.01619 59.9 295.63 efficiency 112 13.73009 1.748721 8.32 19.51 warranty 112 6.568482 2.934662 2 17.57 lifespan 112 0.8011161 0.0061406 0.76 0.82 weight 112 17.40527 2.866048 10.81 25.77 length 112 1582.45 137.9457 1200 1896.32 width 112 880.0682 115.3215 600 1130.99 silicon 112 2163.482 665.9488 1180 3330 chinese 112 0.3125 0.4655956 0 1 production 112 585.8254 773.3039 5 3840 globprod 112 19207.41 14802.13 1199 39987
Notes: (i) “chinese” indicates a dummy for Chinese manufacturer or not. (ii) Production refers to firm total production in year i. (iii) globprod indicates global production in year i.
In order to highlight the evolution of the solar panel industry, I summarize in table 3.3 the arithmetic means of the average performing solar panel calculated per firm. Although this table does not reflect best-in-class technologies and nor that of newly introduced panels, it should conservatively capture the spirit of industry-wide progression. The solar panel industry does not seem to have the same relentless velocity of technological innovation that has been witnessed from personal computers and microchips (Berndt et al, 1995; Jorgenson, 2001), but its technological performance is surely steadily improving.
Table 3.3: Mean values of average performing panels per firm, 2004-2013
Year% N% Cost% Price% Power% Efficiency% Warranty% Lifespan% Weight% Length% Width%
Over a timespan of ten years, prices have been drastically reduced, falling from more than $5/Watt in 2004 to less than $1/Watt in 2013. Although costs have mutually fallen rapidly, the price decline is certainly also related to the increase in Power performance, which has risen by more than 100 Watts. Efficiency shows less rapid advances, but nevertheless has steadily improved by more than 29% over the timespan researched. Furthermore, solar panels appear to have become more reliable as witnessed by the considerable increase in warranty years, while the variable on longevity (lifespan) reveals little action.
Whereas all these variables have moved in the expected direction, the variables on size dimensions (length and width) and weight show no such trend. At first sight that may be surprising, as panels have considerably decreased in size since their early-day introduction, but on closer inspection there are some easy explanations. First of all, it is good to recall that small-size panels below 8kg have been treated as outliers because these often concern differentiated products with very different characteristics, and thus have been discarded from the dataset. Secondly, since solar panels are meant to cover areas such as rooftops (in case of residential installations) or fields (in case of grid installations), their sizes have shrunk to dimensions that are determined by their use. Much like innovation on a microprocessor in a computer device, the innovation for a solar panel mainly comes from the solar cell, which has experienced size reductions. Thus, while there are technologies unfolding to integrate solar cells in small products such as roof tiles, these concern distinct products that have been left out of the dataset.
Another industry development that is worth looking into is the disparity between cost and price evolution. Whereas a surge in polysilicon prices kept cost reductions low in 2008, the more recent severe polysilicon price spikes in 2011 has not prevented solar panel manufacturers from rapid cost reductions. On average, owing to their cost advantage, mark-ups have returned in 2013 for Chinese manufacturers but not yet for the Rest-of-World manufacturers (see figure 3.2, appendix). In order to shed some light on the difference between Chinese versus Rest-of-World manufacturers, figure 3.3 (appendix) summarizes their corresponding cost developments. It is clear that Chinese manufacturers have enjoyed significant cost advantages over their Western counterparts and have contributed to the fierce competition in the industry.
4. Econometric Methodology
Constructing reliable pure price indexes is a challenge that statisticians are continuously faced with. One particular concern is to adjust prices properly for quality change that has occurred over a specific time period (Berndt, 1996). While many products are fairly consistent in their characteristics over time, some products exhibit rapid transformations in their specifications. Often, these products are technological in nature. Solar panels evolve in such fashion, and deflating solar power prices by controlling for quality changes requires specific econometric methodologies.
completely identical in order to prevent a loss of data. Either way, the matched model procedure is not very accurate and often found to understate performance (Berndt, 1996).21 A better method to deal with the complexity of technological change is provided by hedonic price regression estimations, as accurately explained by Berndt (1996). By estimating parameters of the specific characteristics of a product to predict the according price, one can deduct an indication of the quality that is derived from that specific characteristic of the product. As of such, hedonic regression analysis provides a way forward to decompose a solar panel into its constituent characteristics, thus measuring the quality-adjusted price we are looking for (Berndt, 1996). To explain how a quality-adjusted price index can be derived from hedonic regression analysis, consider the following simplification of a hedonic regression model for solar panels with three characteristics for the years 2004 to 2006:
(1) ln Pi = α0 + α1D2005,i + α2D2006,i + β1EFi + β2WRi + β3WT + εi
where the α’s and β’s are unknown parameters to be estimated, EF denotes efficiency, WR denotes warranty, WT denotes weight and ln Pi is the natural logarithm of the panel price. The
error term is captured by εi and assumed to be normally distributed with mean zero and
variance σ2. Automatically, the intercept α0 belongs to the first year of our observations, in
this case 2004. As a result, dummy variables are not required for the first year, and are only included for years 2005 and 2006. It is here that dummy variables are of particular interest to our regression. Since in multiple regression estimations the least squares estimate of a parameter essentially captures the change in the left-hand side dependent variable, it holds the effects of all other parameters fixed. In other words, by estimating the coefficients on time dummy variables, one can very naturally derive a price change while holding quality fixed. Thus, we estimate the change in the natural logarithm of the price between years 2004 and 2005 as follows (Berndt, 1996):
(2) ln p̂2005 - ln p̂2004 = !1
In a very elegant way, equation (2) shows that the quality adjusted logarithmic price change for 2005 is equal to the estimated dummy coefficient for 2005. As a result, if we normalize the price with 2004 as base year (1.00), the quality-adjusted price index for 2005 and 2006 can be computed by exponentiating their respective dummy coefficients (Berndt, 1996):
(3) !"# ! = !!!"
where QAP represents the quality-adjusted price index, t denotes the year, e is an exponent and !t is the estimated dummy variable coefficient in year t.
Thus, by treating solar panels as aggregations of characteristics, hedonic price regressions offer a very elegant and concise way forward to calculating quality-adjusted prices. Despite the advantages of this method, some remarks are in order on potential econometric concerns that might arise. A first disturbance indicated by Berndt (1996) is the presence of heteroscedasticity in the error term that can potentially infect the estimated standard errors. A second issue is related to the choice of variables, as omitted variables that are not perfectly (un)correlated to the included variables can result in biased estimates of the parameters. For !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
21 To the best of my knowledge I have not been able to find data suitable for the construction of matched model
reliable results, both issues require close examination. With that in mind, an extended equation (1) provides the econometric model for the quality-adjusted price index construction: (4) ln Pit = β1 + !!2Xit + !!3Dit + εit
where ln Pi denotes the natural logarithmic price of solar panel i in year t, Σiβ2Xit summarizes the continuous variables with panel-i specific characteristics, Σiβ2Dit summarizes the dichotomous variables on Chinese and years 2005-2013, while εit signifies the error term. From this unbalanced dataset, I pool observations to estimate the yearly dummy variable parameters with log-linear OLS regressions. As just touched upon, it is relevant to report the outcome of diagnostic checks. A plot of the logarithmic histogram of prices reveals that there are no issues with normality. Additionally, from an insignificant Jarque-Bera test one can accept the null hypothesis that there is no heteroscedasticity present in the error term influencing the estimated standard errors. There are, however, concerns with respect to multicollinearity. Unsurprisingly, size as measured by length and width is highly correlated with weight, while silicon prices are highly correlated with (global) production values. Intuitively, the rise of the solar industry has pushed up silicon prices. For reliably regression estimates, I must omit variables that display such high correlations.
5. Results
This section turns to the discussion of regression results, which are presented in table 5.1. While the differences in regression results between normal standard errors (1) and robust standard errors (2) are minimal, striking is the very high explanatory power of the model, as the R-squared indicates a goodness-of-fit of around 85%, yielding a highly significant model overall.
Table 5.1: Regression results and parameter estimates per year
(1) (2) (3)
VARIABLES lprice lprice
(robust s.e.) lprice (robust s.e.) efficiency 0.142*** 0.142*** 0.116*** (0.0262) (0.0330) (0.0311) warranty -0.0144 -0.0144 (0.0192) (0.0191) lifespan -2.407 -2.407 (5.434) (5.461) weight 0.0258* 0.0258 (0.0152) (0.0174) chinese -0.268*** -0.268*** -0.282*** (0.0670) (0.0581) (0.0478) production -9.64e-05* -9.64e-05
dum2008 -0.460** -0.460** -0.294** (0.188) (0.218) (0.134) dum2009 -0.896*** -0.896*** -0.733*** (0.194) (0.210) (0.126) dum2010 -1.170*** -1.170*** -1.047*** (0.204) (0.215) (0.103) dum2011 -1.424*** -1.424*** -1.348*** (0.223) (0.235) (0.111) dum2012 -1.936*** -1.936*** -1.887*** (0.241) (0.273) (0.141) dum2013 -2.176*** -2.176*** -2.226*** (0.278) (0.294) (0.135) Constant 1.577 1.577 -0.356 (4.247) (4.181) (0.399) Observations 112 112 112 R-squared Significance 0.853 0.000 0.853 0.000 0.826 0.000
Note: Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
In line with expectations, the technical characteristic efficiency is highly statistically significant and considered its positive sign has an increasing effect on value and price. Thus it confirms expectations that efficiency is a crucial element of solar panel quality. Conversely, no such significance is found for the other panel characteristics, which all have insignificant t-statistics. Although that is a surprising result when bearing in mind the considerable increase in warranty on solar panels over the years, it could be expected that weight, which is highly correlated with size, would not be a significant factor in explaining solar panel prices. Recall here that the size of a solar panel is determined by its function. A further surprising result is the insignificance of the variable production, which is a proxy for economies of scale and was found to be highly correlated to silicon prices. Lastly, before turning to the year dummy coefficients, the regression results confirm industry reports that Chinese manufacturers have a significant cost advantage, thus allowing them to put additional downward pressure on industry prices. In regression (3), I return to results by only including the significant variables. Although this slightly reduces the coefficients of the variables, it also reduces the standard error estimates and does not produce any other noticeable differences with significance levels remaining equal.
A critical review of the dummy coefficients for the years 2005 to 2013 implies that not only their signs are negative as one would expect, but secondly that all years from 2009 to 2013 are highly statistically significant. Yet, it also raises the suspicion that between 2007 and 2008, a structural change has occurred considering the rising significance level. To test for such difference I conduct a Chow-test, which does not confirm this expectation given its probability of 0.107 (F=2.29). However, for the years spanning 2009 to 2013, one can reject the null hypothesis that these years are equivalent, and conclude that there is a significant price reduction over time.
quality-adjusted price for solar panels declined each year from 2004 to 2013. Secondly, whereas the Chow-test did not reveal a significant difference between the years 2007 and 2008, the annual price change shows a considerable fall-back between these years. A logical explanation for this blip may be the surging price of silicon that peaked in 2008 and again in 2011. Thirdly, although there is quite some disparity in the annual percentage changes, on average solar panel prices witnessed an average annual growth rate of -20.6% over the timespan researched. Quite certainly, in recent years Chinese manufacturers have contributed to this development. Astonishingly, the quality-adjusted price index reveals that a 2004-equivalent solar panel was valued in 2013 at roughly 11% of its original 2004 price. A plot of the quality-adjusted price decline can be found in the appendix, figure 5.1.
Table 5.2: Quality-adjusted Price Index for Solar Panels, 2004-2013
Year Coefficient Estimated (Antilogarithm) Price Index
Annual change (%) 2004 1.0000 2005 - 0.062406 0.9395 - 6.1 2006 - 0.297488 0.7426 - 21.0 2007 - 0.425985 0.6531 - 12.1 2008 - 0.460250 0.6311 - 3.4 2009 - 0.896316 0.4080 - 35.4 2010 - 1.170445 0.3102 - 24.0 2011 - 1.424240 0.2407 - 22.4 2012 - 1.935999 0.1442 - 40.1 2013 AAGR - 2.175627 (2004 – 2013) 0.1135 - 21.3 - 20.6
Notes: (i) 2004 is taken as base year. (ii) AAGR gives the annual average growth rate over the period 2004 to 2013.
6. Summary, discussion and areas for future research
Following the unparalleled developments of information technology that have had fundamental impact on the productivity of business practices, a consensus has emerged that technological innovation takes centre stage in driving 21st century economic growth. However, the assessment of the economic potential of such technological products, given the pace at which innovation develops, requires a profound understanding of the interaction between prices and quality change. One of these technologies is solar power, which recently has been identified by several industry analysts and economists as having disruptive potential for economic growth. Indeed, solar prices have been declining rapidly while technological characteristics have improved, and as a result solar power currently is on the verge of cracking price competitiveness of conventional energy sources. Krugman even referred to this development as resembling a “Moore’s-law” trajectory.
Tour et al., 2013), the purpose of this paper has been to push literature into unchartered territories by mapping the price decline of solar panel prices using a quality-adjusted price index. In doing so I have followed the example of work by Berndt et al. (1995) and Jorgenson (2001), who have conducted similar studies with regard to the price decline of information technology.
Central to the construction of a quality-adjusted price index is the application of a hedonic pricing model. Using a dataset built from scratch, I have used estimated parameters of hedonic regressions to assess the pure price change of solar panels by holding fixed the decomposed technological characteristics of the product. Highly significant regression results from the hedonic model suggest that over a recent time period (2004-2013), solar panels have experienced some thorough technological change. While Chinese manufacturers are found to have put downward pressure on industry prices, quality changes in terms of increased efficiency rates and power generation capabilities have been at the forefront of driving technological progression for solar panels. Surprisingly, no evidence was found in favour of the theory of economies of scale.
Price index results reveal that on average, quality-adjusted prices are found to have declined by 20.6% per year over the timespan researched. Comparably, the average annual growth rate for normal prices is -17.8%. Although this implies that the quality-adjusted price reflects a difference of a relatively modest 2.8% per year, it is relevant to consider the following. Firstly, normal prices to an extent already are quality-adjusted, as a price quotation of $/W implies that prices are not stated in nominal terms but are adjusted for the power of the system. Secondly, over the time period researched, the discrepancy between the quality-adjusted price and the normal price is 25,8% in 2013, a reasonably large gap. That is a relevant result when noting that the quality-adjusted price in 2013 is found to have already cracked the $0,60 price barrier (often stated as the cost-competitiveness barrier), while that is not the case for the unadjusted price ($0,58 versus $0,79). Therefore it may be concluded that taking into account quality changes has impact on assessing the economic viability of solar power.
The findings in this paper bear several implications for statisticians, policy makers and business analysts around the world. Firstly, on a statistical note the deflation of prices according to quality change and technological progress is crucial in correctly measuring GDP levels and economic growth. Since the solar business has turned into a multi-billion dollar business, it is important that statisticians are able to observe the quality-adjusted price of solar power. Secondly, business analysts and politicians should be aware that solar power’s current economic viability is in fact higher when taking into account such quality-adjustments. In stimulating the industry, it is vital that governments are aware of the profoundly negative effects of unstable government policy with respect to solar power. After decades of reliance on government subsidies for survival there are signs that solar power is becoming viable of its own, certainly when new 25%-40% efficiency technologies materialize, but stable policies are needed for the industry’s development. Thus, in production countries such as China, Germany and the U.S. it is important that policies stimulate investment and technological innovation, while governments in sun-rich developing countries should recognize the impressive potential that solar power current-day provides.
extended, and it could also be estimated more reliably by using weighted-averages according to sales data per specific solar panel. Since the solar industry has become much more transparent over the last few years, new solar panel databases will much more reliably track both prices and technical specifications, thus preventing the need to compile a dataset from multiple sources. Solar panels are heterogeneous products, and the technical characteristics I have been able to include should be extended by more elaborate performance specifications. A further note is in order with respect to the econometric method used in constructing the quality-adjusted price index. Firstly, I have not been able to appropriately estimate price mark-ups with respect to firm size. Secondly, to the best of my knowledge, there was no data available yet that could serve for a matched model procedure to verify results. Thirdly, despite the fact that hedonic regression analysis provides a very concise way of price index construction, there are more diverse econometric methods that may additionally be used to verify the established results. Lastly, a promising line of future research is to take into account quality-adjusted prices when analysing energy price competitiveness in terms of levelized cost of energy, which would reflect an even brighter future for solar power.
References
Aanesen, K., Heck S. and D. Pinner (2012). ‘Solar Power: Darkest before Dawn’. McKinsey
on Sustainability & Resource Productivity, McKinsey & Company.
Benthem, A.V., Gillingham, K.T. and J. Sweeney. (2008) ‘Learning by doing and Optimal Solar Policy in California’. The Energy Journal, 29: 131–152.
Berndt, E.R., Z. Griliches and N.J. Rappaport (1995). ‘Econometric estimates of prices indexes for personal computers in the 1990’s’. Journal of Econometrics, 68: 243-68. Berndt, E.R. (1996). ‘The practice of Econometrics: Classic and Contemporary’. Addison-
Wesley Publishing Company, ISBN-13: 978-0201499001.
Chien, J.C. and N. Lior (2011). ‘Concentrating solar thermal power as a viable alternative in China’s Electricity supply’. Energy Policy, 39(12): 7622-36.
De la Tour, A., Glachant, M. and Y. Ménière (2013). ‘Predicting the costs of photovoltaic solar modules in 2020 using experience curve models’. Journal of Energy, 62: 341-48. EIA - US Energy Information Administration, International Energy Outlook 2014,
http://www.eia.gov/forecasts/ieo/more_overview.cfm, accessed 11-09-2014.
European Commission (2014) - Strategic Energy Technologies Information System (SETIS), ‘Solar photovoltaic’. Website:
http://setis.ec.europa.eu/technologies/Solar-photovoltaic/info, accessed 30-09-2014.
EPIA – Masson, G., S. Orlandi and M. Rekinger (2014). “Global Market Outlook for Photovoltaics 2014-2018”. European Photovoltaic Industry Association (EPIA). Frankel, D., Ostrowski, K. and D. Pinner (2014). ‘The disruptive potential of solar power’.
McKinsey Quarterly, McKinsey & Company.
Gauché, P., Brent, A.C. and T.W. von Backström (2014). ‘Concentrating solar power: Improving electricity cost and security of supply, and other economic benefits’.
Development Southern Africa, 31(5): 692-710.
Gordon, R.J. (2014). ‘The Demise of U.S. Economic Growth: Restatement, Rebuttal and Reflections’. NBER working paper, no. 19895, issued February 2014.
Haley, U.C.V. and D.A. Schuler (2012). ‘Government policy and firm strategy in the Solar Photovoltaic Industry’. California Management Review, 54(1)
Jarmin, R.S. (1994). ‘Learning by doing and competition in the early rayon industry’. Journal
Jorgenson, D. (2001). ‘Information Technology and the U.S. Economy’. The American
Economic Review, 91(1).
Jorgenson, D. and K.J. Stiroh (2000). ‘Raising the Speed Limit: US Economic Growth in the Information Age’. OECD Economics Working paper, no. 261.
Hutchby, J.A. (2013). ‘A “Moore’s Law”-like approach to roadmapping photovoltaic technologies’. Renewable and Sustainable Energy Reviews, 29: 883-890.
Kanallos, M. (2011). ‘Is There A Moore’s Law for Solar?’ Forbes Green Tech, accessed 14- 05-2015, available via http://www.forbes.com/sites/michaelkanellos/2011/11/09/is-there-a-moores-law-for-solar/
Krugman, P. (2011). ‘Here comes the Sun’. New York Times,
http://www.nytimes.com/2011/11/07/opinion/krugman-here-comes-solar-energy.html?_r=0, accessed 11-04-2015.
Lorenz, P., Pinner, D. and T. Seitz (2008). ‘The Economics of Solar Power’. McKinsey
Quarterly, McKinsey & Company.
NREL - National Renewable Energy Laboratory, “Solar Photovoltaic Technology Basics”. Website: http://www.nrel.gov/learning/re_photovoltaics.html, accessed 29-09-2014. Masson, G., Orlandi, S. and M. Rekinger (2014). ‘Global Market Outlook for photovoltaics
2014-2018’. European Photovoltaic Industry Association (EPIA).
McKinsey Global Institute, Manyika, J., M. Chui, J. Bughin, R. Dobbs, P. Bisson and A. Marrs (2013). ‘Disruptive technologies: Advances that will transform life, business and the
global economy’. McKinsey & Company Global Institute.
Ming, Z., Shaojie, O., Hui, S. and G. Yujian (2014). ‘Is the “Sun” still hot in China? The study of the present situation, problems and trends of the photovoltaic industry in China’. Renewable and Sustainable Energy Reviews, 43: 1224-237.
Moore, G.E. (1965). ‘Cramming more components onto integrated circuits’. Electronics
magazine, 38.
Naam, R. (2011). ‘Smaller, Cheaper, Faster: Does Moore’s Law apply to solar cells?’
Scientific American, Guest Blog.
Nemet, G. (2006). ‘Beyond the learning-curve: factors influencing cost reductions in photovoltaics’. Energy Policy, 34 (17): 3218-3232.
Neubacher, A. (2012). ‘Solar Subsidy Sinkhole: Re-Evaluating Germany’s Blind Faith in the Sun.’ Spiegel Online International, 3. Also available on:
http://www.spiegel.de/international/germany/solar-subsidy-sinkhole-re-evaluating-germany-s-blind-faith-in-the-sun-a-809439.html
Ondraczek, J. (2014). ‘Are we there yet? Improving solar PV economics and power planning in developing countries: The case of Kenya’. Renewable and Sustainable Energy Reviews, 30: 604-15.
Pietzcker, R.C., Stetter, D., Manger, S. and G. Luderer (2014). ‘Using the sun to decarbonize the power sector: The economic potential of photovoltaics and
concentrating solar power.’ Journal of Applied Energy, 135: 704-720.
Pillai, U. (2014). ‘Drivers of Cost Reduction in Solar Photovoltaics’. Preprint submitted to
Energy Economics.
Pillai, U. and J. McLaughlin (2013). ‘A model of competition in the solar panel industry’.
Energy Economics, 40: 32-39.
Rogers, H. (2007). ‘Current Thinking’. New York Times,
http://www.nytimes.com/2007/06/03/magazine/03wwln-essay-t.html, accessed 11-09-2014.
Saga, T. (2010). ‘Advances in crystalline silicon solar cell technology for industrial mass production.’ NPG Asia Materials, 2: 96-102.
http://www.seia.org/policy/solar-technology/photovoltaic-solar-electric, accessed 29-09-2014.
SEIA 2013 - ‘Solar Market Insight Report 2013 Year in Review’. US Solar Energy Industries
Association (SEIA), website:
http://www.seia.org/research-resources/solar-market-insight-report-2013-year-review, accessed 09-02-2015.
Shrimali, G and E. Baker (2011). ‘Optimal Feed-in Tariff Schedules’. IEEE Transactions on
Engineering Management, 99: 1–13.
Solangi, K.H., Islam, M.R., Saidur, R., Rahim, N.A. and H. Fayaz (2011). ‘A review on global solar energy policy’. Renewable and Sustainable Energy Reviews, 15(4): 2149-163.
Solow, R.M. (1987). ‘We’d better watch out’. New York Times (July 12th) Book Review, 36.
Wesoff, E. (2015). ‘A setback for SolarWorld and a trade case win for Chinese manufacturers’. Greentech Media, website:
http://www.greentechmedia.com/articles/read/A-Setback-for-SolarWorld-and-a-Trade-Case-Win-for-Chinese-Solar-Manufacture, accessed 13-02-2015.
Yu, Y., Song, Y. and H. Bao (2012). ‘Why did the price of solar PV Si feedstock fluctuate so wildly in 2004-2009?’ Energy Policy, 49: 572-585.
Zhang, S., Andrews-Speed, P. and M. Ji (2014). ‘The erratic path of the low-carbon Transition in China: Evolution of Solar PV policy’. Energy Policy, 67: 903-12. Zhao, X., Wan, G. and Y. Yang (2014). ‘The turning point of solar photovoltaic industry in
China: will it come?’ Renewable Energy and Sustainable Energy Reviews, 41: 178-88. Zipp, K. (2013). ‘What Are Solar Panels Made Of?’ Solar Power World,
http://www.solarpowerworldonline.com/2013/05/what-are-solar-panels-made-of/, accessed 30-09-2014.
Appendix
Figure 2.1: Global Installed solar capacity, 2004-2013
Notes: (i) Figure displays the development of global installed capacity per region over time measured as MW. (ii) ROW is “rest-of-world”, MEA is Middle-East and Africa, APAC is the Asia-Pacific region.
Source: EPIA Global Outlook 2014-2018.
Figure 2.2: Residential system cost-composition: US & Germany (2013)
Notes: * Permitting, Inspection and interconnection costs; ** Including installer margins, legal fees, professional fees, financial transaction costs, operation and maintenance costs, production guarantee and warranty. Source: This graph belongs to Rocky Mountain Institute, supported by SunShot (US Department of Energy). Website: http://www.rmi.org/simple, accessed 11-02-2015.
Figure 2.3: Simplified graphical depiction of the solar industry value chain activities
Sources: Ming et al. (2014); Haley & Schuler (2011); Platzer (2012); Green Rhino Energy, website: http://www.greenrhinoenergy.com/solar/technologies/pv_valuechain.php, accessed 10-02-2015
Upstream
• Production of raw materials (mostly polysilicon) • Slicing of polysilicon into ingots
Midstream
• Production of wafers out of ingots • Wafer-cutting into solar cells
• Module production (panel formation, applying protection using rubbers and glass, production of aluminium frame)
Downstream
• Wholesaler
Figure 2.4: US silicon price trend (2000-2013)
Notes: (i) Unit value is the value in actual U.S. dollars of 1 metric ton (t) of silicon apparent consumption. Source: US Geological Survey Data Series.
Figure 3.1: Cost scatterplot between all solar panel manufacturers
Notes: (i) vertical axis measured as $/W.
Figure 3.2: Disparity between Average Cost and Price (2004-2013)
Notes: (i) vertical axis measured as $/W. (ii) Calculations based on geometric means.
Figure 3.3 Average cost development: Chinese vs. RoW manufacturers
Notes: (i) vertical axis measured as $/W. (ii) Calculations based on arithmetic means. (iii) Years 2004 and 2005 are not included as there is only one observation.
Figure 5.1: Index of Quality-adjusted price development (geometric means)
Notes: (i) vertical axis displays the price index with 2004 as base year. (ii) Price calculations are based on geometric means. (iii) Price index is based on antilogarithms as calculated in table 5.2.
Table 2.1: Global installed solar capacity (2003-2013) YEAR / Region 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 ROW 964 993 1003 1108 1150 1226 1306 1590 2098 2098 2098 MEA n/a 1 1 1 2 3 25 80 205 570 953 China 52 62 70 80 100 140 300 800 3300 6800 18600 Americas 102 163 246 355 522 828 1328 2410 4590 8365 13727 APAC 916 1198 1502 1827 2098 2628 3373 4951 7513 12159 21992 Europe 601 1306 2291 3289 5312 11020 16854 30505 52764 70513 81488 Total 2635 3723 5112 6660 9183 15844 23185 40336 70469 100504 138856 Growth (%) 27% 41% 37% 30% 38% 73% 46% 74% 75% 43% 38%
Notes: (i) ROW indicates rest-of-world, APAC indicates the Asia-Pacific region, MEA indicates Middle-East and Africa. (ii) Source: EPIA Global Outlook 2014-2018.
Table 2.3: Solar Photovoltaic Module Production by Top 10 Companies in 2013
Rank Company Production Megawatts
1 Yingli Green Energy 2.622
2 Trina Solar 2.560 3 Canadian Solar 2.020 4 First Solar 1.628 5 JA Solar 1.252 6 Jinko Solar 1.215 7 Kyocera 1.200 8 Flextronics 1.058 9 Hanwha-SolarOne 1.050 10 Solar Frontier 995 World Total 39.987
Source: Compiled by Earth Policy Institute from GTM Research, PV Cell Module Production Data, electronic database, updated June 2014.
Table 3.1: Variable Overview
Variable Definition Function
price Average module price ($/Watt) Dependent variable
power Peak power in Watts Explanatory variable (quality)
efficiency Sunlight conversion efficiency (%) Explanatory variable (quality)
warranty Warranty in years Explanatory variable (reliability)
lifespan Percentage of power guarantee after 25 years Explanatory variable (lifespan)
weight Weight in kg Explanatory variable (size)
length Length in mm Explanatory variable (size)
width Width in mm Explanatory variable (size)
silicon $US per metric ton of silicon consumption Control variable
chinese Dummy variable on Chinese origin Dummy variable
production Total firm production (MW) Explanatory variable (economies of scale)
globprod Total global production (MW) Control variable (learning curve)
Years Year dummies Dummy variables
