Commodity Market Financialisation: A Closer Look at the Evidence

Commodity Market Financiaisation: A Coser Look at the Evidence Aexandra Dwyer, James Hooway and Michee Wright* There is some debate about whether financia investors have caused excessive increases in the eve and voatiity of commodity prices. These investors are viewed by some as being ess concerned with fundamentas than traditiona market participants and hence impeding the price discovery process that is, they are destabiising specuators or noise traders. This artice discusses the reationship between the futures markets for commodities (where financia investors are most active), and the spot markets. It then argues that the evidence does not support the hypothesis that financiaisation has been the main driver of commodity price deveopments in the 2s. Introduction The past decade has seen a sharp increase in the eve and voatiity of commodity prices (Graph 1). This has occurred aongside a sharp increase in commodity demand from emerging market economies, but aso in parae with a rapid increase in both commodity derivatives trading and financia investor activity in commodity markets (Domanski and Heath 27; Dwyer, Gardner and Wiiams 211). That is, commodity markets have aso become somewhat more ike financia markets. This has given rise to considerabe interest in the factors driving commodity prices in particuar the extent to which they have refected fundamenta determinants of demand and suppy versus the growing financiaisation of commodity markets. 1 If the decisions of financia specuators refect informed views about fundamentas, financiaisation can pay a beneficia price discovery roe. However, if financia specuators base their decisions on expectations of future price changes in the absence of fundamenta reasons to do so such as noise or momentum trading behaviour specuation * The authors are from Internationa Department. 1 For a summary, see G2 Study Group on Commodities (211, pp 32 33). Index 2 15 1 5 Graph 1 Commodity Prices* 25 average = 1 1992 1996 2 * IMF A Primary Commodities Index Source: IMF 24 28 212 coud be destabiising (see, for exampe, Franke and Rose (29)). Financia investors are generay most active in futures markets, rather than spot markets, as they do not want to take deivery of the physica commodity, which is expensive to store and to finance. Instead, the roe of financia investors is to act on informed views on the prospects for suppy and demand as we as to be paid to take on the commodity price risk that producers, and to a esser degree consumers, wish to hedge. There are two broad channes Index 2 15 1 5 BULLETIN MARCH QUARTER 212 65

through which commodity futures markets can affect the production and consumption decisions of participants in spot markets: (i) they aow firms to hedge their exposures to movements in spot prices, thereby smoothing their consumption expenditure and/or production cash fows over time and owering the cost of capita; and (ii) they provide a potentia source of infuence over spot prices. If the soe function of futures markets was to provide hedging services to producers and consumers, the wefare impications woud be unambiguousy positive. But if specuation in futures markets causes futures prices to diverge from physica suppy and demand fundamentas, this coud have a distortionary effect on spot prices. In considering this issue, we start by discussing the reationship between spot and futures prices from a theoretica perspective, before considering some of the empirica evidence. Overa, we concude that there is no cear evidence that the financiaisation of commodity markets has had a pervasive effect on commodity prices; instead, the evidence is consistent with fundamenta suppy and demand factors remaining the key determinants of commodity prices. The Reationship between Futures and Spot Prices in Theory The spot price is the cash price paid for the immediate deivery of a physica commodity, whereas the futures price is the price of a standardised exchange-traded contract to purchase/se a specific quantity of a commodity for deivery at a specified future date. In contrast to spot markets, investors in futures markets generay do not actuay participate in the physica deivery of the commodity; instead they ro over their contracts to the next futures contract if they wish to maintain their exposure. This is because physica deivery of the commodity gives rise to storage and financing costs, with itte offsetting benefit to a financia investor from actuay having a physica hoding of, for exampe, soybeans or natura gas. Graph 2 shows the reationship between 1-1 3-3 -6 Graph 2 Spread between Spot and Futures Prices* Weeky % Soybeans US natura gas % % Auminium God -6 22 27 212 22 27 * A positive spread indicates the spot price is above the front-month futures price Source: Boomberg spot and (front-month) futures contract prices over time for soybeans, US natura gas, auminium and god, each of which have reasonaby arge and active futures markets. 2 The theoretica reationship between futures prices and spot prices is based on a no-arbitrage condition. 3 This says that consumers and producers shoud remain indifferent between buying and seing the physica commodity at today s spot price, and entering into a futures contract that woud aow them to buy and se the commodity at a specified ater date at today s futures price. In practice, financing constraints coud imit this process to some extent. Assuming that the commodity is storabe and that (we-informed) participants are abe to freey access both the spot and futures markets (i.e. there are no financing or institutiona constraints), then an unexpected increase in the futures price woud, a ese equa, aow agents to profit from buying the commodity today at the (reativey ow) spot price, and seing it in the future at the (reativey high) futures price. This woud then pace upward pressure 2 The front-month contract for a given commodity refers to the futures contract with the nearest expiry date; it is generay the most iquid futures contract and has the smaest spread to the spot price. 3 Aso, institutiona factors may create a cose reationship between futures and spot prices in some markets, independenty of any direct arbitrage reationship. For exampe, spot or contract prices may be set mechanicay with respect to futures prices. 3-3 3 212 % -3 66 RESERVE BANK OF AUSTRALIA

on the spot price and/or downward pressure on the futures price unti the no-arbitrage condition was restored. Importanty, however, the no-arbitrage condition does not impy that the futures price shoud equa the spot price, or that a given change in the futures price wi be accommodated by a proportionate change in the spot price. This is because the arbitrage reationship aso takes into account some underying differences between physica commodities and futures contracts, which may themseves vary over time. First, there is an opportunity cost associated with buying and hoding the physica commodity, as opposed to entering into a contract to purchase the commodity at a future date and earning interest on the funds set aside for this future purchase in the meantime. This opportunity cost, which is captured by the foregone interest rate, acts to reduce (increase) the return from buying (seing) the physica commodity at the spot price reative to entering into a futures contract. Second, hodings of physica commodities incur storage costs whereas futures contracts do not. Storage costs act to reduce (increase) the return from buying (seing) the physica commodity at the spot price reative to entering into a futures contract. Third, there is a convenience yied from hoding the physica commodity, which is the benefit of having assured access to the commodity in the event of a suppy disruption. This acts to increase (reduce) the return from buying (seing) the physica commodity at the spot price reative to entering into a futures contract. The no-arbitrage condition describing the reationship between futures and spot prices can be represented by the foowing equation: 4 F t = S t e (r+c y)(t t) (1) 4 This equation is a variant of Hoteing s rue, which states that in a competitive market, the price of an non-renewabe resource (net of margina costs) wi increase in ine with the interest rate. Where: F t and S t are the futures and spot prices at time t; r is the risk-free annua interest rate; c is the cost of storing the physica commodity; y is the convenience yied earned from hoding the physica commodity; and T is the maturity date for the futures contract. 5 This basic mode captures the cost of freey avaiabe finance, but can be extended to account for financing constraints and/or risk aversion by incorporating a risk premium. 6 It can be seen from this equation that the futures price wi ony be cose to the spot price if the net impact of the interest rate, storage costs and convenience yied (r+c y) is very sma, or the futures contract reates to a very near deivery date (so that (T t) is very sma). Simiary, changes in futures and spot prices need not be proportionate, depending on how these other variabes change. For exampe, if an increase in the futures price encourages a buid-up of inventories, storage costs may aso rise (as warehouse space becomes scarce) and the convenience yied may fa (as the benefits of physicay hoding a more abundant commodity diminish). The observed increase in the spot price in response to the higher futures price woud then be smaer than otherwise. The no-arbitrage condition aso does not specify whether the spot or futures price adjusts in response to an unanticipated change in one of the variabes in the equation. If, for exampe, there is an unexpected increase in the futures price, the no-arbitrage condition coud be restored by: the futures price subsequenty faing again; the spot price rising; or some combination of the two. In practice, this wi depend on the extent to which the unexpected change in the futures price is perceived to refect 5 In Equation (1), c and y are expressed as proportions of the spot price for iustrative purposes. 6 A risk premium woud be expected to drive a wedge between futures and spot prices, particuary if investors are highy risk averse. Intuitivey, this risk premium can be thought of as the compensation required by financia specuators to participate in the futures market, since their participation is not derived from a need to hedge an exposure in the physica market. Adding this risk premium term (which can be positive or negative depending on whether the desired net position of commodity producers and consumers in the futures market is short or ong) woud ater the form of Equation (1) to: F t = S t e (r+c+ρ y)(t t), where ρ is the risk premium. BULLETIN MARCH QUARTER 212 67

a genuine change in fundamentas, as we as the time horizons of participants in the spot and futures markets. If an increase in the futures price is viewed as reveaing genuiney new information about fundamentas, firms that suppy the physica commodity to the spot market wi have an incentive to buid inventories, whie firms that demand the physica commodity wi have an incentive to stockpie purchases for future use. This shoud create excess demand for the commodity in the spot market at the current price, thereby pushing the spot price up unti the no-arbitrage condition is restored. In this scenario, futures prices woud ony distort spot prices if there are information faiures that is, if participants in the spot market mistake specuative price deveopments for genuine price discovery. However, if an increase in the futures price is not considered to revea any genuiney new information about fundamentas, the response of firms in the spot market (and we-informed investors in the futures market) wi depend on their views about how ong the apparent bubbe-ike conditions wi be sustained, and how ong they are wiing to hod their positions. 7 Such a situation coud arise, for exampe, due to the infuence of so-caed noise or momentum traders, who are either ess we-informed than other market participants, or who activey choose to ignore fundamentas (Sheifer and Summers 199; Reichsfed and Roache 211). If the deviation from fundamentas is considered temporary, firms that suppy the physica commodity to the spot market wi have an incentive to increase their short positions in the futures market (i.e. enter into agreements to se the commodity at a future date at the reativey high futures price, rather than at the (ower) expected spot price). At 7 Specuative price movements coud aso occur in spot markets. However, such instances are ikey to be reativey isoated, as uninformed financia investors (who have no underying physica demand for commodities) are, in genera, ess ikey to participate in spot markets (where they wi incur storage and financing costs without an offsetting convenience yied). Whie market manipuation by informed participants in spot markets may aso be possibe, this is unreated to the financiaisation of commodity markets. the same time, firms that demand the commodity in the futures market wi have an incentive to reduce their ong positions in the futures market. This shoud pace downward pressure on the futures price, to the point where the no-arbitrage condition is restored. Aternativey, participants in the spot market may suspect that a rise in the futures price which is not justified by fundamentas coud be sustained for exampe, due to herding behaviour among noise traders. In this case, producing firms may be tempted to withhod suppy to the spot market (in expectation that the higher futures prices wi transate into higher spot prices) and reduce their short futures positions (which provide insurance against fas in the spot price). At the same time, consuming firms wi have an incentive to stockpie the spot commodity for future use and increase their ong futures positions (which provide insurance against increases in the spot price). Other, better-informed, financia specuators may aso be encouraged to bet on future price increases in order to book short-term profits. This process coud simutaneousy drive spot and futures prices higher, and even further from the price impied by fundamentas. Whie it may be reasonabe to expect fundamentas to eventuay reassert themseves, so-caed rationa bubbes coud nevertheess act to distort spot and futures prices for some time. The Reationship between Futures and Spot Prices in Practice With that background in mind, it is usefu to examine how these reationships pay out in practice. To this end, we perform Granger causaity tests to examine the empirica reationship between daiy changes in spot and (front-month) futures prices that is, whether changes in one price systematicay precede changes in the other for a range of individua 68 RESERVE BANK OF AUSTRALIA

commodities. 8 In the context of asking how financia specuators coud infuence prices, there are four possibe outcomes of these tests, each with different impications: If changes in futures prices are found to Granger-cause changes in spot prices, this suggests that price discovery is occurring in the futures market. This coud indicate that the futures market tends to absorb news about changes to fundamentas more quicky than the spot market. A ess benign interpretation coud be that specuative deveopments in futures prices are distorting spot prices (at east temporariy). If changes in spot prices are found to Granger-cause changes in futures prices, this suggests that price discovery is occurring in the spot market. In this case, any specuation-driven changes in futures prices are unikey to distort spot prices. If we find evidence of bi-directiona Granger causaity (i.e. changes in futures prices Granger-cause changes in spot prices and changes in spot prices Granger-cause changes in futures prices) this indicates that spot and futures prices are jointy determined. This coud indicate either that there are a arge number of participants with access to both markets (such that perceived news is simutaneousy refected in both the futures and spot markets) or that there are institutiona factors which enforce a cose mechanica reationship between the two markets. Lasty, if no Granger-causa reationships are detected, this may suggest that spot and futures markets are sufficienty segmented to prevent 8 More formay, the percentage change in the spot price is regressed on agged changes in both the spot price and the futures price. If the estimated coefficients on the agged changes in the futures price are found to be jointy statisticay significant (using a Wad test) then changes in the futures price wi be said to Granger-cause changes in the spot price. A simiar regression is then run for the percentage change in the futures price. Bi-directiona Granger causaity occurs when both variabes are found to Granger-cause each other (i.e. they are jointy determined). If a variabe does not hep predict the other, no Granger causaity is said to exist. arbitrage from occurring, and therefore that deveopments in one market are unikey to affect the other. Aternativey, arbitrage may sti hod, with the Granger-causa reationships existing ony on an intraday basis or adjustment occurring primariy through changes in other variabes (e.g. through storage costs or the convenience yied). Granger causaity tests are estimated for 1 commodities, covering four commodity casses base metas, agricuture, energy, and precious metas over a sampe period from 1997 to 211. Detais of the price measures used are shown in Appendix A. We aso perform the tests over two sub-periods 1997 to 22 and 23 to 211 to determine if the reationships between spot and futures prices have changed as commodity futures markets have become much arger. The tests are conducted using a standard GARCH (1,1) mode for ag engths ranging from 1 to 1 days. 9 Tabe 1 presents the resuts of the Granger causaity tests for ag engths of 1, 5 and 1 days (which are generay representative of the resuts obtained using other ag engths). The resuts for base metas (auminium, copper, nicke and zinc) are mixed, but there is itte evidence of a consistent one-way Granger-causa reationship from futures prices to spot prices (i.e. that changes in futures prices systematicay precede changes in spot prices). Instead, we find evidence of a bi-directiona Granger-causa reationship for copper and nicke, but amost no evidence of a Granger-causa reationship in either direction for zinc or auminium. The bi-directiona Granger-causa reationships between futures and spot prices for copper and nicke suggest that these prices are typicay jointy determined and are therefore ikey to be anchored 9 A GARCH mode is used because high-frequency financia time series typicay exhibit voatiity custering, whereby arge changes in a variabe tend to be foowed by other arge changes and sma changes tend to be foowed by other sma changes. GARCH modes expicity estimate this reationship and in so doing are abe to estimate more accurate standard errors than an ordinary east squares approach. The (1,1) specification for the mode was seected based on the evidence in Hansen and Lunde (25) and the Akaike and Schwarz Bayesian Information Criteria. BULLETIN MARCH QUARTER 212 69

Tabe 1: Test Resuts for Granger Causaity between Spot and Futures Prices (a) 1997 211 (b) 1 day 5 days 1 days Auminium None None None Copper Both Both Both Nicke Both Both None Zinc None None None Corn Futures Spot Futures Spot Futures Spot Soybeans Futures Spot Futures Spot Futures Spot Wheat Futures Spot Futures Spot Futures Spot US natura gas Futures Spot Futures Spot Futures Spot God Spot Futures Spot Futures Spot Futures Siver Both Both Spot Futures (c) 23 211 1 day 5 days 1 days Auminium None None None Copper Both Both Both Nicke Both Both Futures Spot (d) Zinc None None None Corn Both Futures Spot (c) None Soybeans Futures Spot Futures Spot Futures Spot Wheat Futures Spot Futures Spot None US natura gas Futures Spot Futures Spot Futures Spot God Spot Futures Spot Futures Both Siver Spot Futures Spot Futures Spot Futures (a) Resuts are statisticay significant at the 5 per cent eve, except where otherwise indicated; London Meta Exchange (LME) prices are used for base metas, Chicago Board of Trade (CBOT) prices are used for agricutura commodities (b) Juy 1997 to December 211 (c) Bi-directiona Granger-causa reationship at the 1 per cent eve of significance (d) Bi-directiona Granger-causa reationship at the 1 per cent eve of significance for ags up to and incuding nine days Sources: Boomberg; authors cacuations to a common set of fundamentas. On the other hand, whie the absence of any Granger-causa reationship between changes in spot and futures prices for auminium and zinc coud suggest that there are barriers to arbitrage between the two markets, it is arguaby more ikey that futures and spot price adjustments are occurring on an intraday basis, which is not captured by the daiy frequency of our data. It is aso possibe that some adjustment occurs through other factors, such as storage and/or financing costs. The resuts for the agricutura commodities (corn, soybeans and wheat) are much more uniform, with strong evidence that daiy changes in futures prices Granger-cause daiy changes in spot prices. This is not surprising, as spot markets for agricutura commodities tend to be reativey fragmented (i.e. they consist of a reativey arge number of producers with speciaist oca knowedge). These resuts aso hod in the 23 211 sub-sampe, except at onger ag engths for corn and wheat where there no onger appears to be a Granger-causa reationship in either direction. These findings indicate that, for 7 RESERVE BANK OF AUSTRALIA

these agricutura commodities, deveopments in futures prices have a bearing on spot prices. For US natura gas, we aso find strong evidence that daiy changes in futures prices Granger-cause daiy changes in spot prices. Oi prices are deiberatey excuded from the Granger causaity anaysis as there are certain institutiona features of the oi market which compicate the reationship between spot and futures prices. In particuar, there is arguaby no independent benchmark spot market for oi (see Fattouh (211) for a more detaied discussion of the features of the oi market). So, for exampe, for West Texas Intermediate (WTI) oi, the benchmark (Cushing crude oi) spot price trades at parity to the front-month futures price for a but a 3-day deivery scheduing period that commences when the current front-month futures contract expires. For precious metas (god and siver) we find some evidence that spot prices Granger-cause futures prices, particuary over the most recent period. God spot prices Granger-cause god futures prices (athough with some evidence of bi-directionaity at onger ag engths in the more recent period). For siver, there is a argey bi-directiona Grangercausa reationship over the fu sampe period, but over the 23 211 period, spot prices are found to Granger-cause futures prices. There does, however, appear to be some weak evidence of a return to bi-directiona Granger-causaity during the rapid run-up in siver prices between mid 21 and end 211, suggesting that deveopments in siver futures prices did have an effect on spot prices during this so-caed bubbe episode. More generay, the apparent infuence of precious metas spot prices on futures prices is ikey to be reated to the reativey arge and iquid nature of spot markets for these commodities, which in turn refects their unique status as financia assets with reativey ow storage costs. Reated to this, the growth in physicay backed commodity exchange-traded products for precious metas may aso be a factor, as these products require investment in the underying physica commodity at the spot price (Kosev and Wiiams 211). Puing a this together then, it seems the reationship between spot and futures prices is a compex one, varying across commodities, sometimes within commodity casses, and aso over time. There is evidence for agricutura commodities and US natura gas that changes in futures prices ead those in spot markets. If futures prices for these commodities refect fundamentas, these markets can be viewed as being wefare enhancing, with the participation of financia specuators adding to the iquidity of these markets and improving price discovery. However, if there is evidence of specuation in these futures markets by noise or momentum traders, this has the potentia to distort the corresponding spot prices, with adverse consequences for the rea economy. Consequenty, to distinguish between these competing views on the roe of financia specuators, it is important to evauate the evidence on the reationship between futures prices and macroeconomic fundamentas. Are Futures Price Deveopments Consistent with Fundamentas? Previous Reserve Bank work has found that, in genera, the arge increase in the number of financia investors in commodity markets over the past decade has not significanty atered price dynamics (see Dwyer et a (211)). The main pieces of evidence in support of this view are that: (i) price increases have been just as arge (if not arger) for some key commodities that do not have we-deveoped financia markets as for those that do (Graph 3); (ii) there has been substantia variation in the price behaviour of individua commodities, even among those that have arge, active derivatives markets (such as natura gas and oi); (iii) the recent increase in the correation between commodity prices and other financia prices, such as equities which is commony cited as evidence that financia specuators are affecting prices is in fact not that unusua by onger-run historica standards, with previous BULLETIN MARCH QUARTER 212 71

COMMODITY MARKET FINANCIALISATION: A CLOSER LOOK AT THE EVIDENCE episodes of increased correation occurring prior to the recent infux of financia investors into commodity markets (Graph 4); and (iv) there does not appear to have been the arge increase in commodity inventories that we woud expect to accompany specuation-driven price rises (as discussed in the earier section on the theoretica reationship between futures and spot prices). In this artice, we present two further pieces of anaysis which suggest that, in genera, deveopments in futures prices have been Graph 3 Commodity Prices Percentage change between January 23 and February 212 Iron ore Coking coa Copper Therma coa God Brent oi LNG Rice WTI oi Sugar Corn Soybeans Wheat Cotton Oats Auminium US natura gas -2 2 4 6 8 % n Sma futures markets n Large futures markets* * Number of contracts traded in 21 greater than 25 miion Sources: ABS; Boomberg; Energy Pubishing; gobacoal; RBA ρ.75.5.25. -.25 Graph 4 Commodity and Equity Price Correations Five-year roing window, seected time periods 1962 212, daiy* 192 212, monthy** -.5 -.5 1962 1987 212 197 21 * Date abes correspond to the first day of the year; correation between daiy percentage changes in Thomson Reuters/Jefferies CRB and S&P 5 indices ** Date abes correspond to the first month of the year; correation between monthy percentage changes in Thomson Reuters/Jefferies CRB and S&P 5 indices Sources: Goba Financia Data; RBA ρ.75.5.25. -.25 consistent with fundamentas. First, we show that the reationship between commodity prices and the goba output gap over the past decade is broady in ine with that seen over a onger time horizon (athough the omission of suppy-side factors makes it difficut to draw any firm concusions about the reationship between commodity prices and fundamentas based on the output gap aone). Second, we use principa component anaysis to show that since 23, individua commodity prices have been driven primariy by a singe common factor, which appears to be reated to macroeconomic fundamentas. Commodity prices and the goba output gap It has been argued that the goba output gap is an important determinant of the cycica behaviour of commodity prices, since commodities are used as an input to production (and typicay it takes some time for commodity suppy to respond to changes in demand). As shown in Inamura et a (211), there appears to be some evidence of this, with a broad co-movement over time between the goba output gap (measured as the difference between actua and potentia goba GDP) 1 and various commodity price indices (reproduced in Graph 5). 11 Proponents of this view suggest that increased financia investment in commodity markets over the past decade has resuted in an upward shift in the reationship between commodity prices and the goba output gap. Abstracting from suppy factors, the intuition here is that financiaisation constitutes a source of increased demand for commodities which is unreated to macroeconomic fundamentas (as captured by the output gap). Graph 6 pots the reationship between rea commodity prices and the goba output gap from 1971. There does 1 Whie industria production may be a more reevant measure of goba activity for this purpose, we use GDP in order to assess the caims made in previous research. Goba GDP is measured using purchasing power parity exchange rates and potentia output is cacuated using the Hodrick Prescott fiter (λ = 1 6). 11 For information on the differences between seected commodity futures price indices, refer to RBA (211). 72 RESERVE BANK OF AUSTRALIA

Index 3 2 1 1995 Graph 5 Graph 6 The Goba Output Gap and Commodity Prices* and the Goba Output Gap Commodity Prices Yeary 1971 1994, quartery 1995 211 DJ-UBSCI* (LHS) 1999 S&P GSCI* (LHS) 23 Goba output gap (RHS) 27 2-2 -4 211 * Spot returns measures of the S&P Godman Sachs Commodity Index (GSCI) and the Dow Jones-UBS Commodity Index (DJ-UBSCI) defated by the US CPI; 2 average = 1 Sources: Boomberg; Goba Financia Data; IMF; RBA % Rea commodity prices - index 4 3 2 1 1979 1986 23 211** 1971 1978 1995 22 1987 1994-4 -2 2 4 Goba output gap - % * Spot returns measure of the S&P GSCI defated by the US CPI; 2 average = 1 ** Data to September quarter 211 Sources: Goba Financia Data; IMF; RBA indeed appear to have been an upward shift in the reationship between rea commodity prices and the goba output gap between 1995 22 and 23 211, consistent with the financiaisation hypothesis. However, taking a onger-run historica perspective, it is evident that the reationship observed over the 23 211 period is around average, whereas it is the reationship from 1995 22 (and aso 1987 1994) that ooks unusua. That is, it is the period of ow and faing rea commodity prices during the atter part of the 198s and the 199s that ooks more unusua, rather than the most recent period. As noted above, however, this anaysis omits suppy-side factors, which are aso important determinants of commodity prices. In particuar, suppy conditions were tight in the 197s associated with the oi price shocks but eased in the 198s in response to the earier increase in prices. So, from a onger-run perspective, the reationship between commodity prices and the goba output gap in recent years does not ook unusua. In any event, the omission of suppy-side factors means that any change in this reationship cannot, of itsef, be attributed to the financiaisation of commodity markets in recent years. Principa component anaysis An aternative way to examine the extent to which deveopments in commodity futures prices have been consistent with macroeconomic fundamentas is through principa component anaysis. This statistica technique identifies whether there are common factors driving movements in an underying set of observed variabes, and how important they are, without having to specify what those factors might be. Drawing on this anaysis, together with broader evidence on the drivers of commodity prices (see, for exampe, Connoy and Orsmond (211); Dwyer et a (211)), we can infer the extent to which these common factors are reated to macroeconomic fundamentas. This anaysis was conducted on quartery price changes for a broader set of 2 commodities over two sampe periods: the September quarter 199 to the December quarter 22 and the March quarter 23 to the December quarter 211. 12 By comparing the resuts from these two periods, we can gain 12 The anaysis for the atter period was aso performed over a sighty onger time period (March quarter 2 to the December quarter 211) to test the sensitivity of the resuts to the use of a reativey short time period. The resuts from this exercise were very simiar to those obtained over the shorter period. BULLETIN MARCH QUARTER 212 73

Tabe 2: Principa Component Anaysis of Changes in Commodity Prices Share of variation expained (per cent) Principa component 23:Q1 211:Q4 199:Q3 22:Q4 1 4 23 2 12 14 3 9 11 4 7 8 5 6 7 6 5 7 7 4 6 8 3 5 9 3 4 1 3 3 2 Source: authors cacuations some insights into the effect of financia investment in commodity markets. The resuts suggest that, since 23, one common factor (i.e. the first principa component) has expained 4 per cent of the tota variation in our set of 2 commodity prices in change terms (Tabe 2), with the next most important factor accounting for ony 12 per cent. 13 In eves terms, the first principa component expains amost 7 per cent of the variation since 23. A number of statistica tests indicate that there is ony one significant common factor. 14 The resuts show that the first principa component has become more important over the past decade compared with the 199s, when it ony expained 23 per cent of the variation in commodity prices in change terms (and just under 13 The principa component anaysis is performed using percentage changes in quartery (daiy average) front-month futures prices. The exception to this is the use of LME spot prices for base metas from the start of the sampe period to Juy 1997 due to the unavaiabiity of LME futures prices up unti this time. The resuts of the principa component anaysis aso hod for a (smaer) sampe of spot, rather than futures, prices. 14 The standard Scree test and the criterion deveoped by Otter, Jacobs and den Reijer (211) show that there is one significant common factor, whie the Bai-Ng Pane Information Criteria suggest one or two common factors, depending on which statistic is used (Bai and Ng 22). 4 per cent in eves terms). Moreover, across the various commodities, the first factor oadings (i.e. the correations between changes in the commodity s price and the first principa component) are reasonaby uniform within the recent sub-period (Tabe 3). US natura gas prices are one notabe exception, consistent with the arge (idiosyncratic) increase in suppy associated with the shae gas revoution together with the fact that US natura gas is restricted to the domestic market due to a ack of internationa transportation infrastructure. Agricutura prices aso tend to have somewhat ower oadings on the common factor. This seems ikey to refect the importance of idiosyncratic particuary weather-reated suppy factors in driving futures prices for agricutura commodities. The dominance of the first principa component shows that there has been one major common driver of deveopments in commodity prices, particuary in the post-23 period. This appears ikey to be reated to known macroeconomic deveopments over this period in particuar, the combination of an unexpectedy arge increase in demand for commodities and suggish suppy growth. For exampe, the pair-wise correation between the 74 RESERVE BANK OF AUSTRALIA

Tabe 3: First Factor Loadings for Individua Commodity Prices Correation between price change and first principa component 23:Q1 211:Q4 199:Q3 22:Q4 Auminium.82.61 Copper.79.56 Oats.77.48 Siver.76.12 Brent oi.75.89 WTI oi.72.87 Heating oi.71.82 Corn.66.48 Cotton.64.11 Zinc.63.31 Soybeans.62.21 Lead.61.24 Coffee.55.12 Nicke.54.48 Cocoa.51.23 God.51.15 Wheat.5.43 Rice.46.32 Sugar.39.9 US natura gas.38.33 Source: authors cacuations first principa component and quartery growth in goba industria production is.7 over the recent period. Whie this does not contro for other reevant variabes, such as suppy factors, it is nevertheess broady consistent with the resuts obtained from a more fuy specified mode in Arbati and Vasishtha (212). The aternative hypothesis, which is that financiaisation has been by far the most important infuence on commodity prices, is a much ess pausibe expanation, in arge part because there is no reason to suspect that this has an eement to it that is common across a rather disparate set of commodities, a number of which are not even incuded in the major commodity indices used by financia investors. Concusion Overa, whie financia specuation at times may have exerted some infuence on some commodity prices beyond fundamentas, the avaiabe evidence does not support the hypothesis that financiaisation has been the main driver of commodity price deveopments in the 2s. More generay, the theoretica reationship between commodity futures and spot prices does not impy that changes in futures prices need necessariy ead to changes in spot prices. In practice, this is supported by the resuts of Granger causaity tests, which point to substantia variation across individua commodities. R BULLETIN MARCH QUARTER 212 75

Appendix A Tabe A1: Spot and Futures Prices Used in Granger Causaity Tests and Principa Component Anaysis Commodity Spot price Futures price Agricutura Cocoa (a) na Intercontinenta Exchange US Coffee (a) na Intercontinenta Exchange US Corn United States Department of Agricuture Grain Export Chicago Yeow Number 2 Chicago Board of Trade Cotton (a) na Intercontinenta Exchange US Oats (a) na Chicago Board of Trade Rice (a) na Chicago Board of Trade Soybeans United States Department of Agricuture Yeow Number 1 Chicago Board of Trade Sugar (a) na Intercontinenta Exchange US Wheat Base metas Auminium United States Department of Agricuture Soft Red Winter Number 2 London Meta Exchange Primary Auminium Cash Chicago Board of Trade London Meta Exchange Primary Auminium Copper London Meta Exchange Cash London Meta Exchange Lead (a) na London Meta Exchange Nicke London Meta Exchange Cash London Meta Exchange Zinc London Meta Exchange Cash London Meta Exchange Energy Brent oi (a) na Intercontinenta Exchange Europe Heating oi (a) na New York Mercantie Exchange US natura gas Henry Hub New York Mercantie Exchange WTI oi (a) na New York Mercantie Exchange Precious metas God Boomberg god spot price COMEX Siver Boomberg siver spot price COMEX (a) No spot prices are reported as these commodities were not used in the Granger causaity tests Source: Boomberg 76 RESERVE BANK OF AUSTRALIA

References Arbati E and G Vasishtha (212), Growth in Emerging Market Economies and the Commodity Boom of 23 28: Evidence from Growth Forecast Revisions, Bank of Canada Working Paper No 212-8. Bai J and S Ng (22), Determining the Number of Factors in Approximate Factor Modes, Econometrica, 7(1), pp 191 221. Connoy E and D Orsmond (211), The Mining Industry: From Bust to Boom, RBA Research Discussion Paper No 211-8. Domanski D and A Heath (27), Financia Investors and Commodity Markets, BIS Quartery Review, March, pp 53 67. Dwyer A, G Gardner and T Wiiams (211), Goba Commodity Markets Price Voatiity and Financiaisation, RBA Buetin, June, pp 49 57. Fattouh B (211), An Anatomy of the Crude Oi Pricing System, The Oxford Institute for Energy Studies No 4. Franke JA and AK Rose (29), Determinants of Agricutura and Minera Commodity Prices, in R Fry, C Jones and C Kent (eds), Infation in an Era of Reative Price Shocks, Proceedings of a Conference, Reserve Bank of Austraia, Sydney, pp 9 43. G2 Study Group on Commodities (211), Report of the G2 Study Group on Commodities under the Chairmanship of Mr Hiroshi Nakaso, November. Avaiabe at <http://www.banque-france.fr/fieadmin/user_upoad/ banque_de_france/economie_et_statistiques/tendances_ Regionaes ne_pas_ecraser_/mois_impairs/study_ group_report_fina.pdf>. Hansen PR and A Lunde (25), A Forecast Comparison of Voatiity Modes: Does Anything Beat a GARCH(1,1)?, Journa of Appied Econometrics, 2(7), pp 873 889. Inamura Y, T Kimata, T Kimura and T Muto (211), Recent Surge in Goba Commodity Prices Impact of Financiaization of Commodities and Gobay Accommodative Monetary Conditions, Bank of Japan Review, 211-E-2. Kosev M and T Wiiams (211), Exchange-traded Funds, RBA Buetin, March, pp 51 59. Otter PW, JPAM Jacobs and AHJ den Reijer (211), A Criterion for the Number of Factors in a Data-rich Environment, University of Groningen, unpubished manuscript, February. Avaiabe at <http:// www.eco.rug.n/medewerk/jacobs/jjdownoad/ CriterionForNumberOfFactors_Feb211.pdf>. RBA (Reserve Bank of Austraia) (211), Box A: A Comparison of Commodity Indices, Statement on Monetary Poicy, November, pp 14 16. Reichsfed DA and SK Roache (211), Do Commodity Futures Hep Forecast Spot Prices?, IMF Working Paper WP/11/254. Sheifer A and LH Summers (199), The Noise Trader Approach to Finance, The Journa of Economic Perspectives, 4(2), pp 19 33. BULLETIN MARCH QUARTER 212 77

78 RESERVE BANK OF AUSTRALIA