Is opacity-‐induced minor metal market
volatility a threat to promising green
technologies?
A study of the tellurium market
Fredrik Söderqvist
Master of Science Thesis
Uppsala University Department of Economics Submitted June 7, 2013
This master thesis was written in the spring of 2013 as part of the Uppsala University Master Programme in Economics. I would like to thank Sander de Leeuw at New Boliden AB for the support, inspiration, and data access he has generously granted me, and my supervisor Mikael Bask for his thoughtful guidance and meticulous supervision of this thesis.
For questions, comments or inquiries regarding the content, methods, data or
Abstract
Tellurium is one of the rarest metals in the earth’s crust. Increased demand for cadmium telluride photovoltaic cells along with an opaque pricing and quantity-‐ reporting system, have recently caused high price volatility and a speculative bubble in the tellurium market, resulting in overstocking and depressed prices. In a longer perspective this may be a threat to cadmium telluride photovoltaics as a power-‐generating technology. This master thesis compares how actors may perceive news innovation in the opaque tellurium market compared to the more transparent molybdenum market. A quantitative analysis of industry news reporting on the two metals, combined with a SVAR impulse response analysis, helps me determine which actors and factors exert most influence on spot market prices. In the opaque tellurium market, relatively unreliable proxies of supply and demand are most frequent in the news reporting while having a big impact on prices, whereas the transparent molybdenum market uses more reliable variables – such as futures prices – and transparent supply information, whilst also relying on a frequent stream of dependable proxies to scope market sentiments. My findings lead me to recommend policy makers to implement measures to increase market transparency, which may be accomplished by extending the data-‐sharing regime of the REACH database to minor metal markets. Attempting to limit speculation in minor metal markets is perhaps too blunt a tool to fix an inherent problem of a free exchange-‐pricing mechanism.
Sammanfattning
Tellur är en av de mest sällsynta metallerna på Jorden. Ökad efterfrågan av kadmiumtelluridsolpaneler har nyligen orsakat stor volatilitet på tellurmarknaden. Ett opakt prissättnings-‐och kvantitetsrapporteringssystem har bidragit till att en prisbubbla bildats och spruckit, vilket resulterat i att marknadsaktörer köpt på sig stora lager till höga priser som de sedan inte kunnat sälja vidare. I ett längre perspektiv kan detta innebära begränsningar vid tillverkning av solcellsteknologi baserad på kadmiumtellurid, då ett volatilt pris kan göra nya tellurgruvprojekt alltför riskabla. Denna masteruppsats jämför hur en typisk marknadsaktör kan reagera på prisinnovationer i den opaka tellurmarkanden och den mer transparenta molybdenmarknaden. Metoden består av en kvantitativ analys av facknyheter rörande de två metallerna, varifrån variabler väljs till en SVAR modell med impuls-‐responsanalys. Urvalet av variabler är få och volatila på den opaka tellurmarknaden, medan den mer transparenta molybdenmarknaden har ett större utbud av variabler som kännetecknas av god transparens och relativ förutsägbarhet. Mina slutsatser leder mig till att rekommendera beslutsfattare att vidta åtgärder för att öka tellurmarknadens transparens genom EU-‐samarbetet, förslagsvis genom att göra anonymiserad data från REACH databasen tillgänglig för allmänheten. Samtidigt avråder jag från åtgärder som syftar till att minska spekulation, då implementering av en sådan policy kan bli både dyr och komplicerad.
Key words: Tellurium, Minor Metal, Market Volatility, Market Transparency, Molybdenum, Market Efficiency, REACH, SVAR, Quantitative Analysis, London Metal Exchange.
Table of Contents
IS OPACITY-‐INDUCED MINOR METAL MARKET VOLATILITY A THREAT TO
PROMISING GREEN TECHNOLOGIES? A STUDY OF THE TELLURIUM MARKET ... 1
ABSTRACT ... 4 SAMMANFATTNING ... 4 1. INTRODUCTION ... 6 2. BACKGROUND ... 8 2.1 TELLURIUM ... 8 2.2 TELLURIUM SUPPLY ... 9 2.3 TELLURIUM DEMAND ... 11
2.4 THE TELLURIUM MARKETPLACE ... 12
2.5 THE TELLURIUM MARKET TODAY ... 13
2.6 MOLYBDENUM -‐ A NOT-‐SO MINOR METAL ... 14
2.7 CRITICAL MINOR METALS ... 15
2.8 PREVIOUS STUDIES OF MINOR METAL MARKETS ... 15
3. METHOD: DETERMINING THE PRICE MECHANISMS OF TELLURIUM AND MOLYBDENUM ... 17
3.1 SVAR AND IMPULSE RESPONSE FUNCTIONS ... 17
3.2 QUANTITATIVE ANALYSIS ... 18
4. DATA AND RESULTS ... 22
4.1 SPOT PRICES AND RETURNS ... 22
4.2 QUANTITATIVE ANALYSIS FINDINGS ... 24
4.3 INCORPORATING APPROPRIATE ACTORS AND FACTORS INTO THE SVAR MODEL ... 28
4.3.1 The Yu et al (2012) model on applied on tellurium ... 28
4.3.2 A market-‐ specific tellurium model ... 32
4.3.3 A market-‐specific molybdenum model ... 36
4.4 OTHER FINDINGS FROM THE QUANTITATIVE ANALYSIS ... 40
5. CONCLUSIONS ... 44
REFERENCES ... 46
APPENDIX ... 50
LIST OF ABBREVIATIONS ... 50
VAR AND SVAR FUNCTION DERIVATION ... 51
QUANTITATIVE ANALYSIS CODING EXAMPLE ... 53
COMPLETE STRUCTURAL INNOVATION GRAPHS ... 54
1. Introduction
As photovoltaic (PV) technologies recently reached grid parity without government subsidies in several places and thus becoming a cheaper source of power compared to buying electricity from the power grid1, demand for critical
materials in PV technologies is expected to increase. One of these critical materials is tellurium (Te), a minor metal2 and one of the rarest metals in the
Earth’s crust. Until recently, Te has mainly been used as a machinability-‐ increasing alloying agent in steel manufacturing. The metal’s semi-‐conducting properties – when bound with cadmium to produce Cadmium Telluride (CdTe) – have proven excellent at converting solar radiation into electricity in CdTe PV solar cells. CdTe PV is, as of February 2013, the most efficient technology to harness the power of the sun with regards to costs per watt produced ($/Wp) and conversion efficiency; however, long term CdTe growth may be hemmed by the limited supply of Te and its relative rarity. Despite demand looking positive in the long run, the spot market price for Te tells a conflicting story. Prices have rocketed and fallen in recent years, and thus volatility is very high. To compare the highs and lows; in June 2004 99.99% pure Te cost $31 per kg on the open market, seven years later in June 2011 it cost $430 per kg, and in June 2012 the spot price was only $145 per kg. Like most minor metals, Te is not listed on any commodities bourse, and there exists little reporting of traded quantities. This makes business and long-‐term investment difficult for actors on the market and could threaten future development of CdTe PV production. At a UK House of Commons Science and Technology Committee (2011)-‐ hearing, it was suggested that critical metal market supply-‐information, such as Te supply, should be improved, and measures to limit speculative buying should be considered in order to remedy volatility in minor metal markets.
This thesis is an attempt to determine what causes volatility in the Te metal market. The two main research questions are: which factors, actors, and market institutions have the biggest impact on Te prices, and what does this tell us about the overall trading conditions on the market? The results and methodology could lend conclusions valid to other industry-‐critical, opaquely traded minor metals, and add to the discussion as to what can be done to reduce volatility in these markets. This thesis also contributes to the scientific literature concerning Te supply limitations to CdTe PV, which to my knowledge has not focused on the threat to the future supply of Te that high price volatility may pose.
In order to determine what makes the Te price fluctuate, a SVAR-‐model with impulse response functions is estimated using the same aggregated macroeconomic variables which Yu et al (2012) used to attempt to determine price fluctuations in the photovoltaic silicon feedstock (PVSF) spot market. PVSF is a highly price volatile, critical material in a rival PV solar cell technology. A
1 REneweconomy article UBS: Boom in unsubsidised solar PV flags energy
revolution: http://reneweconomy.com.au/2013/ubs-‐boom-‐in-‐unsubsidised-‐ solar-‐pv-‐flags-‐energy-‐revolution-‐60218 (accessed May 21 2013).
2 A metal included in the Minor Metal Trade Association:
quantitative analysis is then applied to a set of articles published in an industry newspaper, the Metal Bulletin, in order to better select variables that are more market-‐specific. In order to benchmark and better relate the Te results, the same method is applied to molybdenum (Mo), which is a minor metal with similar characteristics and applications as Te. The selection of Mo is mainly motivated by its introduction to the London Metal Exchange (LME) in 2010; a market regime transition that introduced futures contracts, and transparent pricing and quantitative reporting mechanisms.
My findings indicate that market-‐specific actors, factors and institutions amply describe price fluctuations in both the Te and Mo markets, whereas the aggregate macroeconomic variables presented by Yu et al (2012) do not explain price fluctuations well. The quantitative analysis suggests that there are few variables to choose from in the Te market (mainly market specific stock companies). These variables explain price fluctuations quite well, but are not very transparent. On the Mo market there are plenty of proxies of supply, indices, and futures prices that amply explain variation, whilst exhibiting steady information flows of transparent price and quantity reports. From this I advise that measures are taken in the Te market to introduce some of the institutions that help reduce volatility on the Mo market. I deem that the most critical measure would be to improve quantitative transparency in the market, which could be done within the data-‐sharing regime of the REACH framework.
In the second chapter, a background to Te, its supply, demand, marketplace, and market today is given, along with a brief introduction to the Mo market, a definition of minor metals, and a summary of older studies regarding minor metal market information, efficiencies, deficiencies and transparency. In the third chapter, the SVAR model, as presented by Yu et al (2012) is introduced, along with a description of my quantitative analysis. In the fourth chapter, spot prices and returns of Te and Mo are selected. Results from the quantitative analysis are then presented, from which variable selection is made, followed by SVAR and impulse response function results from the Yu et al, Te-‐market specific and Mo-‐ market specific SVAR models. Finally, other findings from the quantitative analysis are presented. In the last chapter I discuss my conclusions and policy recommendations.
2. Background
2.1 Tellurium
Te is an element in the same family as oxygen, sulphur, selenium and polonium. Its abundance on Earth, as displayed in Figure 1, shows that it is one of the nine rarest metals, where seven of these are considered “precious” (Green 2010). Te has semi conducting properties, meaning it has the electrical properties of both a conducting metal and an insulator (Nussbaum 1962). Te supply has traditionally been a by-‐product of copper, lead, and zinc processing, but can also be extracted from gold processing (Green 2009, New Boliden 2011) and is mined as a primary metal on two locations in China, and one in Mexico (USGS, 2013a).
Figure 1 Shows that Te (inside the yellow Rarest “metals”-‐cloud) is one of the 9 rarest metals in the Earth’s crust. Its abundance is similar to that of gold (Au) and platinum (Pt). Source: USGS 2002.
In recent years, an increase in demand for Te has taken place due to a change in the primary industrial usages of the metal. The Selenium Tellurium Development Association (STDA), whose members include most of the world’s major producers of Te, estimates that global distribution by consumption is 40% in solar cells, 30% in thermoelectric and photoelectric copying devices, 15% in metallurgy as an alloying metal, 5% in rubber formulation as a vulcanisation-‐ and acceleration in rubber compounding processes, and 10% in other applications such as in blasting caps and ceramic-‐ and glass pigments (STDA, 2012, USGS 2012a).
The 40% final consumption in photovoltaic cells is due to a recent demand surge that started around the year 2000, when production of CdTe thin PV solar panels increased as a result of technological advancements and government subsidies of PV (Candelise et al, 2011).
2.2 Tellurium supply
Estimating an exact volume of world supply for tellurium is difficult. Many countries and companies do not report their production, while volumes recovered from recycled photoelectric devices is not reported at all (USGS, 2012a). The United States Geological Survey (2013) has chosen to withhold total US-‐output from the public in order to “avoid disclosing company proprietary data”, and due to inaccuracies in the data, have chosen to list world output as N/A since 2006. The British Geological Survey (BGS, 2013) has since 2007 published data on Canadian, American, Peruvian and Japanese-‐produced tonnages of Te, estimating US production at 50 tonnes per year. To add to the inaccuracies, all global production estimates are only based on Te produced from copper anode slimes.3 As Te is not traded on any major bourse, there are no
accounting or reporting requirements – such as those associated with the London Metal Exchange (2013) – and thus traded quantities remain unreported. This means that an estimated BGS (2013) “total world production” (approximately 96 tonnes) as reported by Speirs et al (2011), is much lower than real production, as it omits data from Te-‐producing countries such as Australia, Belgium, Chile, China, Colombia, Germany, India, Kazakhstan, Mexico, the Philippines, and Poland (USGS, 2013a). The most thorough estimate of total world production from copper anode slimes is between 450 and 500 tonnes per year was carried out by the UK consultancy firm Oakdene Hollins (2012).
When discussing future supply of a metal, so-‐called reserves and reserve bases must be taken into account. Reserves are defined by the USGS as the part of the reserve base, which could be economically extracted or produced at a time of determination. Reserve bases are identified sources of a mineral which meet physical and chemical criteria related to current mining practices, and that may one day be extracted economically (USGS 2012a). Reserves reported by the USGS show only reserves of Te bound to copper ores, and are thus an underestimation with regards to real Te reserves. The Oakdene Hollins report (2012) estimate the copper anode slimes reserves to be close to 24 000 tonnes of Te.
Scientific literature concerned with photovoltaic progress has made several attempts to estimate present and future world supply of Te, as CdTe technology will not be a viable power generation technology without a steadily available supply of Te. In a meta-‐study of Te availability, Candelise et al (2011) summarises data from six studies between 1998 and 2009 that estimates future yearly cumulative supply of Te from 128 to 2000 tonnes per year. A common fault in many of these estimates is that they use the above-‐mentioned underestimated USGS data to reach their conclusions. Green (2009) does a further analysis of possible Te that can be extracted from other ores, and so-‐ called Bonanza deposits that mines Te as a primary metal. Hourari et al (2013) is the latest attempt, and looks at future supply from a dynamic perspective, which means that it implicitly takes Te prices and future demand of CdTe into account when estimating future supplied quantities of Te in 2050. The supply is made dynamic by taking other possible final usages of Te into account, as well as
3 A product of electrolysis copper refinement, from which impurities such as Te
including Te which could be extracted when recycling spent CdTe PV units. The study concludes that future Te supply available for CdTe PV production is expected to be slightly lower than in previous studies. These global flows and feedback loops may in the end influence both supply and demand. Figure 2 illustrates how loops of Te supply are determinant for the production of CdTe PV.
Figure 2 The CdTe casual loop diagram, which highlights areas where production costs of producing CdTe PV can be reduced. Source: Houari et al (2013).
Figure 3, the dynamic model, visualises where future sources of Te may come from, and where it may end up.
Figure 3 The system dynamics model where annual Te production plays a big role. Source: Houari et al (2013).
2.3 Tellurium demand
Future demand of Te is dependent on estimates of future technological advancements and production improvement, as well as demand for PV power generation. To measure economic efficiency gains in CdTe technology, an index of USD cost per Watt produced is often used, which enables comparison through time and competing power-‐generating technologies. The latest Cost per Watt
produced estimate by the market leader First Solar, is $0.68/Wp per panel, at
record breaking 20% solar conversion efficiency, making it the most cost-‐ efficient PV technology readily available to the market (First solar 2012). This cost per panel is not the same as cost per PV-‐system or facility, which are generally higher.
A working paper by Speirs et al (2011) gives a clear overview of potential future demand of Te in CdTe PV manufacturing. Future demand of Te is dependent on the above-‐mentioned cost of producing electricity. The working paper shows that the limited future supply of Te should not be a threat to CdTe development, as CdTe PV-‐units will in the future require less Te to produce the same amount of energy. Figure 4 illustrates the content of a CdTe PV thin film cell and how much of it is composed of an active CdTe layer. This layer is expected to decrease in the future through technological progress. Woodhouse et al (2012) have calculated that at a CdTe module produced at $0.70/Wp spends $0.15/Wp on the CdTe active layer, and that future material intensity will decrease from 74 tonnes of Te per GW today, to 17 tonnes per GW in 2020.
Figure 4 Illustration of composition of a CdTe thin film solar cell. The thickness of the Active CdTe is an area believed possible to make thinner, which would decrease future demand for Te. Source: Speirs et al (2011).
supply shortages, which will ultimately lead to higher prices. The previously mentioned papers on the possible limitations on CdTe PV posed by supply shortages make some estimates to a maximum price where power generation would still be profitable, such as Candelise et al (2011), who estimate a maximum spot price of $700/kg of 99.99% Te, and Green (2009) at $800/kg. Woodhouse et al (2012) estimate that at current prices, production in 2020 will be constrained at 10 GW of annual production, which may only be remedied with higher prices that make future mining projects more profitable.
Thus, Te availability ought not to constrict future production of CdTe PV as long as costs for Te do not exceed a certain threshold, and the active CdTe layers in the panels continue to decrease. This thesis attempts to fill a gap in the scientific literature, namely to provide a more robust study of how price mechanisms can affect future Te price scenarios, which has been requested in most of the key literature used in this thesis (Candelise et al 2011, 2012, Green 2009, 2010, and Speirs et al 2011).
2.4 The tellurium marketplace
Te is traded through long-‐term supply contracts and individual trades between large consumers and suppliers. Potential buyers and sellers can list proposed prices on specialist websites, which are then matched. Price quotes usually represent expert estimates of representative prices in trades being executed on a particular day, and not actual traded volumes and prices (Oakdene Hollins 2012). My anonymous source (2013) with good insight in the market adds minor metal conferences and companies’ existing costumer networks as possible forums to meet potential customers. These marketplaces are thus thoroughly opaque to outsiders. The only “open” marketplace I have found is the Chinese trading website Alibaba, where sellers can post advertisements to sell various qualities and quantities of Te.4
Te prices are posted on several trading and market news sites, including the Metal Bulletin, a UK-‐based paper that reports on global non-‐ferrous metals and steel markets (Metal Bulletin 2013a). As tellurium is not traded on any bourse, prices are estimated with the aid of different metal warehouses. Metal Bulletin, which lists many different spot prices of metals and commodities, has done this for many years. The goal is to discover at what level market participants have concluded business, made offers or received bids over a certain time period; usually the period between the last price-‐listing in the paper. After interaction with market actors, Metal Bulletin confirm the transaction with both sides, weigh the price and quantity to other transactions during the time period, and finally post a price listing consisting of a low and high price. They reserve the right to remove any data they consider outliers or discount prices they consider questionable. Metal Bulletin stress that they attempt to engage (and encourage engagement) with all sellers and buyers on the market, irrespective of size, are
4 This market can be accessed by searching for Tellurium on www.alibaba.com
or via the link:
impartial and independent, and do not have any vested commercial interests in pricing of their listed metals. The smallest traded lots taken into consideration when determining the price of Te is 250kg, which recently changed from 500kg (Metal Bulletin 2013b).
Figure 5 illustrates how volatile the spot price of Te is, which is a common trait for many minor metals (Candelise et al 2012).
Figure 5 Te average weekly price from February 6 2006 to February 28 2013. Note: There are no price listings for June 12 and 26, as well as October 2 2009. Source: FOB USA Warehouse (February 4 2006 to June 22 2012) and Metal Bulletin (June 29 2012 to February 22 2013).
2.5 The tellurium market today
In a volatile spot market based on estimates of long-‐term contracts, there may be incentives for actors to ride bubbles for short-‐term profits (Harrison et al 1978, Biasis et al 1998). For example, in June 2011 the price of Te peaked at $430/kg, up from $165/kg in 2009. After the 2011-‐peak, spot prices declined steadily for a year and are stabilised at levels just above $100/kg. This is indicative that the two-‐year 160% increase in price bears the markings of a speculative bubble. A similar phenomenon can be observed for the years 2006 to 2008, when prices more than doubled and then dropped to half its peak value. It has been suggested that these bubbles were initiated by speculative buying of Te under the pretext that the limited supply of the metal would be insufficient to meet future demand (USGS, 2013b). This lead to a hoarding of the material in warehouses, bought at inflated prices. Once the market discovered this, the price rapidly fell, and prices are still depressed, as the stocked Te bought during the bubble has yet been depleted (Oakdene Hollins, 2012). The recent change in minimum reported quantities in the Metal Bulletin from 500kg to 250kg might further be interpreted as an indicator that volumes on the market are currently so low, that making statistical samples of market interactions are difficult at these volumes.
reduced volatility, or else suppliers would find it hard to finance future mining projects of the metal.5
The bubble may be the result of speculative trading on an opaque market that lacks transparent reporting over whom trades what to where, which has resulted in high volatility. In a conference paper, Green (2010) compares Te price fluctuations to those experienced by photovoltaic silicon feedstock (PVSF), which is a material whose price has recently been studied by Yu et al (2011). This observation is discussed further in the method chapter.
In order to compare how an opaquely traded minor metal may differ from a transparently traded minor metal, I make an assessment of the market for molybdenum (Mo), which is traded under a more transparent market regime, and is listed on the London Metal Exchange (LME).
2.6 Molybdenum -‐ a not-‐so minor metal
It is difficult to justify a comparison of the market of one chemical element to another; should chemical characteristics, chemical family, application, or price be used as a basis for comparison? I have chosen to compare Te to Mo for the following reasons: they are both minor metals of similar atomic number (Mo no. 42 and Te no. 52); they are by-‐products of copper production, and thus their supply relies heavily on the extraction and refinement of copper; and they can both be used as steel alloying agents. Finally, Mo was one of two minor metals introduced to the LME in February 2010, which may help to illustrate how a minor metal is traded under the transparent market conditions which were implemented prior to the LME-‐introduction (Oakdene Hollins, 2012).
Mo is a refractory metallic element principally used as an alloying agent in iron, steel, and superalloys to enhance desirable properties such as machinability, toughness, strength and corrosion-‐resistance (USGS, 2012b). These properties, along with it having one of the highest melting points of all the chemical elements, means that Mo has few chemical substitutes. Mo does not exist in nature as a free metal, and is usually found in deposits bound to low-‐grade porphyry-‐molybdenum and copper deposits. The most important ore is molybdenite, and total world supply is roughly composed of half Mo mined as a primary product and half as a by-‐product of copper mining. Final usages of the metal are 24% stainless steel, 16% full alloy steel, 11% tool-‐ and high-‐speed steel, 10% high strength low alloy (HSLA) steel, 9% carbon steel, 6% cast iron, 8% catalysts, 6% metal & alloys, 5% superalloys, and 5% others (Oakdene Hollins, 2012). An interesting development is the relatively small-‐scale application of Mo in CIGS-‐PV6 cells as an electrical conductor, which lends the
metal a small application-‐ connection with the Te market. Data of yearly production and usage of Mo is readily available and indicates a market roughly in balance with regards to supply and demand (IMOA, 2011).
5 “Tellurium price seen in $100-‐150/kg range this year – 5N Plus” by Martin
Hayes, http://www.fastmarkets.com/minor_metals/5nt1 (accessed on March
26, 2013).
Mo spot prices are reported using the same sampling procedure as Te (Metal Bulletin, 2013). Recently, an official cash price was also made available via the LME, which differs slightly in sampling procedure, but will not be used in this thesis due to the limited time span of the data. The main difference between the two metals is there exists a futures market for Mo via the London Metal Exchange (LME, 2013), and thus the spot prices can be seen as a reflection of long-‐term contracts traded transparently on a free market. Although only 6 702 tonnes of Mo had been traded on the bourse between its opening and March 2012, which amounts to approximately 1 % of total estimated traded volumes (Oakdene Hollins, 2012), one can argue that the mere existence of a regulated futures market will reduce volatility (Slade, 1988).
2.7 Critical minor metals
Apart from being considered minor metals, Mo and Te have both been assessed for their criticality by the European Commission (2010). To qualify as a critical material, a raw material must “face high risks with regard to access to it, i.e. high
supply risks or high environmental risks, and be of high economic importance… the likelihood that impediments to access occur is relatively high and impacts for the whole EU economy would be relatively significant.” Many of the materials
considered in the report are minor metals. Although this assessment from 2010 did not qualify Mo or Te as critical materials, the 2011 the Commissions Joint Research Centre (JRC, 2011) added Te to the list due to it being a critical material in strategic energy technologies.
In January 2013 the US Federal Energy Department (2013) followed suit by adding Te to a research hub of critical materials known as the Critical Materials Institute (CMI). The hub mainly focuses on research that reduces supply risks to the metal, which includes making extraction techniques more efficient and reducing the usage in production and manufacturing.
2.8 Previous studies of minor metal markets
Although I have not found any studies on the effects of information transparency on a minor metal market, I have found older papers that are tangent to the subject. The first example is Lee et al (1998), who concludes that increased transparency helps the price discovery process become more efficient, by looking at how the opening of limit order books in the Korean stock exchange in 1992 decreased price volatility and increased liquidity in the stock market.
cost of increased market volatility. Apart from profits, the exchange system has a significant advantage due to its pricing transparency, which means that the transaction price is always true and uniform to all customers. Well functioning institutional rules, such as contract enforcement, where breaches may lead to (a very public) expulsion from the exchange, is another reason why a system shift took place. Slade’s description of a period of transition between two systems of price setting captures the resistance many had (and still have) to future markets; namely that they are inherently risky and bubble-‐inducing. More recent studies have disproved this, and attribute this superstition to a lack of understanding of how transparent futures market actually work (Irwin et al, 2009).
Hallwood (1988) argues that an unregulated exchange market is not as efficient as a regulated one. At the time, copper contracts were traded on the LME, but industry preference meant contracts were often negotiated using LME futures as a benchmark. These prices are by definition less efficient than the LME-‐ negotiated contracts, and caused prices that fluctuated more than actual cyclical demand. According to this argument, the low-‐volume Mo market of today will become less volatile if higher volumes are traded over the LME. Eggert (1991) looked at how prices of more commonly traded metals and commodities fluctuate more compared to consumption of the metal, thus pointing out inefficiencies in the market. The debate focused mainly on whether or not the market could be deemed efficient. The final say in the debate was the disproval of efficiency by Sephton and Cochrane (1990). Although debating whether or not a market could be deemed efficient was a frequently debated topic at the time, proving or disproving a specific market’s efficiency may be considered an antiquated discussion today. However, these discussions revolved around a proposed paradigm shift in pricing systems, and need to be read from that perspective.
This thesis does not focus on the nature of the Efficient Market Hypothesis per se, but acknowledges that more information and transparency both lead to a more efficient market and reduced price volatility. I conclude that the results from these early studies carry little validity in today’s markets where global news have a much more instantaneous effect of markets, nor does their topic of discussion add much to current academic debate. In the following chapter a method is selected to determine how markets react to availability of information, which may differ depending on the efficiency of the market.
3. Method: Determining the price mechanisms of tellurium
and molybdenum
3.1 SVAR and impulse response functions
Robust long-‐term prognostics of price scenarios on a volatile market are difficult to design, and as with all forecasts under high volatility, results are often imprecise and should merely be seen as best guesses of future scenarios. Still, if one can better understand what makes the market tick, decisions regarding future investments may become better informed. This is what Yu, Song, and Bao (2012) attempt to do by modelling real price fluctuations of PVSF, which is the primary component of a PV technology rival to CdTe. This is done by studying impulse response functions on number of variables using a Structural Vector Autoregressive Model (SVAR) that includes (p) periods of lag.
𝐴!𝑧! = 𝛼!+ 𝐴!𝑧!!! ! !!! + 𝜀!
where 𝑧! is a 𝑘×1-‐vector of the 𝑘 variables that are to be studied; 𝛼 is a constant 𝑘×1-‐vector; 𝐴! is the time-‐invariant 𝑘×𝑘-‐ matrix where the main diagonal terms are set to 1. 𝜀! is the 𝑘×1 error term, which satisfies the assumptions E 𝜀! = 0, or every error term has mean zero; E 𝜀!𝜀!′ = Σ, or the contemporaneous matrix of error terms is Σ (a 𝑘×𝑘 positive-‐semidefinite matrix); and E 𝜀!𝜀!!! = 0, meaning for every non-‐zero 𝑘, there is no correlation across time, or more specifically, no serial correlation in individual terms across time.
A SVAR model imposes restrictions on the response of underlying Vector Autoregressive (VAR)-‐variables, meaning one can include assumed inter-‐ variable causality, from which impulse response functions can be calculated using OLS estimation. More information on derivation and assumptions of the VAR and SVAR models are found in the Appendix.
For 𝑒! = 𝐴!!! 𝜀!, we can incorporates the causality assumptions for each model
into the 𝐴!!!-‐matrix. The optimal number of lags (p) is then determined using the
Akaike Information Criterion (AIC).
In the Yu et al model, 𝑧! = (𝑒𝑢𝑟𝑜!, 𝑛𝑎𝑡!, 𝑜𝑖𝑙!, 𝑎𝑔𝑔!, 𝑐𝑜𝑛!, 𝑠𝑝𝑜𝑡!), where the lagged
variables 𝑒𝑢𝑟𝑜! represents euro-‐to-‐dollar exchange rate, 𝑛𝑎𝑡! and 𝑜𝑖𝑙! the price of natural gas and oil, 𝑎𝑔𝑔! real economic activity, and 𝑐𝑜𝑛! and 𝑠𝑝𝑜𝑡! represents
contract-‐ and spot prices of PVSF, all expressed in logs. I use the same assumptions as Yu, Song, and Bao, which can be read in Section 3.1.2 in their article. These assumptions are translated into the equation below, where the diagonal 𝑎!!= 𝑎!!= ⋯ = 𝑎!! = 1 by construction.
𝑒! ≡ 𝑒!!"# 𝑒!!"# 𝑒!!"# 𝑒!!"! 𝑒!!" = 𝑎!! 0 0 0 0 0 𝑎!! 𝑎!" 0 0 𝑎!" 𝑎!" 𝑎!! 0 0 𝑎!" 𝑎!" 𝑎!" 𝑎!! 0 𝑎!" 𝑎!" 𝑎!" 𝑎!" 𝑎!! 𝜀!!"# 𝜀!!"# 𝜀!!"# 𝜀!!"! 𝜀!!"
This thesis attempts to develop this model further by placing greater emphasis on variable selection. The above Yu et al (2012)-‐ variables are selected to best capture macroeconomic impacts on the market. As PVSF is a critical component of a rival technology, the Yu et al-‐ variables and restrictions should work just as well for the Te market. However, I believe market specific shocks may better capture fluctuations on a specific market via market spillover effects (Morales, 2008). To do this, inter-‐variable causality in the SVAR-‐model has to be explicitly stated, and then translated into the (𝐴!!! )-‐causality assumption matrix as is done
above. The error term matrix (Σ) is estimated separately and indicates if the error term assumptions are fulfilled.7 This thesis only considers short-‐term
causality shocks to the Te and Mo prices, which means that Te and Mo spot price will not have an effect on other market variables in the short run.8
From the SVAR model, structural impulse response functions and Cholesky accumulated response functions are then calculated. The structural impulse response function gives an indication of how a response variable reacts to a one standard deviation shock from an impulse variable. The Cholesky function is a measure of how an accumulated one standard deviation shock to an impulse variable affects the mean square error of a response variable, expressed as a fraction of the response variable’s total mean square error. This gives a measure of how much a shock of the impulse variable affects a response variable’s deviation from its mean, or more explicitly: its volatility.
This thesis is a continuation of the discussion called for by Yu et al regarding variable selection, as they did not achieve significant results in their paper. In some sense, it is also an attempt to validate the appropriateness of using a SVAR-‐ model to assess how different variables impact critical minor materials. Apart from applying the Yu et al macroeconomic variables to the Te spot price, this thesis investigates which variables more specific to the Te and Mo markets are appropriate, which is established using quantitative analysis methodology described in the next section.
3.2 Quantitative analysis
Selecting reliable market-‐specific variables presents some difficulties to a layman not familiar with a market. In order to determine which factors and actors may be deemed most important in a market, a content analysis is carried
7 All models and estimations are done using STATA 12. The causality
assumptions of the 𝐴!!!-‐matrix is input as the A-‐matrix, and the standard
assumptions for the Σ-‐matrix is input as the B-‐matrix.
8 Estimating long-‐run impulse response functions could capture these causalities.
out on a set of articles published in the Metal Bulletin. The coding scheme is designed with reliability, validity, accuracy, and precision in mind, using method established in Neuendorf (2002). The methodological inspiration partly comes from Tetlock (2007), which uses a simple quantitative analysis on a popular Wall
Street Journal column to study how the media and stock prices interact, and
Tetlock et al (2008), that looks at how linguistic qualities in firm-‐specific news reporting may predict a firm’s accounting earnings and stock returns; or more specifically, how the market has a tendency to underreact to the usage of words that may reveal negative sentiments on returns and earnings. My approach is different to these studies, and focuses more on determining if and how a news innovation is expected to cause a price change, and who is the catalyst of the event.
The selection of SVAR-‐model variables takes frequency of actor-‐ and factor-‐ mentions, market mechanisms, and other insights from the quantitative analysis into account. Actors and factors can either have an effect on supply, such as stocks of mining companies, or demand, such as stocks of consumers of the metals. If possible, effects of actors and factors are quantified using their respective stock prices, and relevant factor indices.
The articles are collected from the Metal Bulletin news archive by searching for the terms tellurium –“MB NON-‐FERROUS PRICE CHANGE” and molybdenum –“MB
NON-‐FERROUS PRICE CHANGE”. The –“MB NON-‐FERROUS…”-‐term excludes so-‐
called price-‐update articles, which are not proper news articles, but listings of daily price changes. All articles from February 20 2010 until February 28 2013 are pasted into word documents and imported into excel-‐spread sheets where the coding scheme is inserted at the top of each sheet.
The decoding of the articles is done in six steps. The first step determines whether the news article is price-‐pertinent; or can the described event in the article theoretically change the price of the metal? Examples of non-‐pertinent articles are those that do not directly deal with the supply or demand of Te or Mo, such as those dealing with Te as an impurity in steel scrap. Examples of pertinent topics include business reports of increased production, changes in market conditions, opening of new mines, or reporting on changes in trade barriers. Articles may also be deemed pertinent if the content is deemed relevant to the research question of my thesis.
If the article is deemed pertinent, the next step is to determine the general topic of the article, which is best described as a one-‐sentence description of the article’s effect on a metal price. This is done with the purpose of improving referencing ability, so the description does not need to be consistent with how previous topics are coded.