THE GLOBAL URANIUM MARKET

Duke Cole Reserve Bank of Australia

November 2015

ABSTRACT By applying methods common to the commodities and energy literature to uranium for the first time, we test whether the uranium market acts differently to other markets. In theory, the uranium market faces potentially significant distortionary effects because of a strong regulatory framework, competition from unconventional secondary supply, and the idiosyncratic nature of nuclear power generation. We find evidence of clear similarities with other commodities, but also discover evidence of strange behaviour. The nature of the uranium market is analysed in three ways. First, by comparing the cyclical behaviour of the uranium price to a broad cross-section of commodities using the Bry-Boschan algorithm. Relative to the rest of the sample, uranium experiences short, sharp booms and long, deep slumps. Second, the relationship between regional uranium prices and privately calculated price indicators are tested using cointegration; these results suggest that the price indicators are important reference prices in lieu of a spot market for uranium. Third, bivariate cointegration tests are conducted between uranium, and both its coproducts and energy substitutes. Only crude oil exhibits a (weak) cointegrating relationship with uranium, which challenges the result of past supply-demand work on the uranium market.

This paper is based on work completed while at the University of Western Australia. Views expressed in this paper are those of the author and do not necessarily reflect the views of the Reserve Bank of Australia. Use of any results from this paper should clearly attribute the work to the author and not to the Reserve Bank of Australia. The author is solely responsible for any errors.

1.

INTRODUCTION1

At first glance the global uranium market appears to be different when compared to other commodities. Overshadowed by its military legacy, the market faces a strict regulatory environment and unpredictable competition from secondary supply. Additionally, uranium demand is wholly dependent on the slow construction of costly, capital-intensive nuclear power plants. But do these factors materially affect the dynamics of the uranium market? Uranium’s primary civilian use is in nuclear power, though a little is also used to make medical radioisotopes. Nuclear power is the largest single source of low-carbon electricity globally. Thirty countries include it in their energy mix, with 13 using it to generate more than a quarter of their electricity. Uranium makes up a small proportion of the total cost of electricity relative to other forms: 31 per cent, compared to 78 per cent for coal. The largest three producers of uranium (Kazakhstan, Canada and Australia) are responsible for around 60 per cent of output. Since 1990, consumption has consistently outpaced mined production (Figure 1.1). This gap (15 per cent in 2010) has been filled by various sources of secondary supply (OECD/IAEA, 2012). Secondary supply includes downblended nuclear weapons (the highly-enriched uranium is diluted with natural uranium) and large inventories that were built up during early years of uranium production, due to overly optimistic consumption forecasts and government interference in the enrichment market (Figure 1.2; Owen (1992)). Figure 1.1: Uranium production and consumption, kilotonnes kT Production 60

40

Consumption

20

0 1970

1975

1980

1985

1990

1995

2000

2005

2010

Sources: BREE, OECD/IAEA.

Contracts in the uranium market are determined bilaterally as no formal exchange exists, and long-term contracts appear to dominate the market. A futures market was established in 2007, though it is still quite illiquid. According to the 1968 Treaty on the Non-Proliferation of Nuclear Weapons, all trading nations except the Nuclear Weapon States (NWS) must sign a Comprehensive Safeguards Agreement which allows the International Atomic Energy Agency to track and verify nuclear materials.2 This – and other similar rules – strictly regulate the transfer of uranium, potentially posing a substantial barrier in the market. The existing literature on the economics of uranium primarily focuses on estimating supply-demand models. 3 The most recent analysis by Kahouli (2011) emphasises the links between uranium, and competing fuels and coproducts. He finds – contrary to earlier research – that the uranium price has a significant negative effect on uranium demand, and is also significant in determining supply. Kahouli also finds that coal is the only fuel which is positively correlated with the uranium price. Finally he suggests that the gold price has a significant 1 This paper is based on work completed with the help of financial support from the UWA Business School. I thank Professor Ken Clements and the Economics staff for their comments and feedback. 2 The five NWS are China, France, Russia, the United Kingdom and the United States. 3 Ahmed (1979), Owen (1985), Amavilah (1994), Trieu et al. (1994), Amavilah (1995).

1

positive effect on uranium supply, implying that an increase in the price of gold will increase the supply of uranium mined as a gold coproduct. These structural models are limited by a lack of high-frequency data and hampered by the structural changes because of the socio-political and regulatory influence on the market. For example, inventory levels are generally modelled with simple dummy variables, despite changes in inventories being advanced as a key reason for fluctuations in the uranium price. We take a different approach by examining the price dynamics of uranium using methodologies common to the commodities and energy literature. Uranium is generally excluded from broader energy and commodity research. 4 Monthly price data are available for the last forty years, covering most of the history of peaceful uranium consumption. The price we use is an average of indicators calculated by private firms. Uranium can be characterised in a similar manner to other fuels and commodities despite the structural barriers present in the market. This paper contributes to the literature on uranium and commodities in three ways: •

The Bry-Boschan algorithm is applied to uranium for the first time, and its cyclical behaviour is compared to a broad cross-section of commodities. The Bry-Boschan algorithm formalises graphical methods of determining turning points. Uranium displays relatively long, deep slumps punctuated by short, sharp booms. The cointegration of regional uranium prices and privately calculated price indicators is tested for the first time. We find that the regional prices move together with the price indicators. Therefore they substitute for a competitively determined price in lieu of a spot market. Finally we use cointegration to test uranium’s relationship with both its coproducts and other fuels. The uranium price is not integrated with that of natural gas, coal, nor any coproducts, and is only weakly integrated with the crude oil price.

• •

This paper continues as follows. Chapter 2 discusses uranium’s price cycles. The next two chapters contain the cointegration analysis; first applied to regional uranium prices (Chapter 3), then to test for relationships between uranium and its coproducts and energy substitutes (Chapter 4). The final chapter concludes. Figure 1.2: Comparison of actual uranium requirements and forecast uranium requirements kT 300

250 1979

1982

200 1976

1977

150 1973 2007 100 1991

2011 2003

1969

50

Actual Uranium Requirements

1986 0 1970

1975

1980

1985

1990

1995

2000

2005

2010

2015

2020

2025

Notes: Uranium requirements are a proxy for uranium consumption. Dotted lines represent forecasts of uranium requirements published in Uranium Resources Production and Demand in the year shown. Where high and low forecasts were provided, the average was taken. Sources: BREE, OECD/IAEA. 4

As far as we are aware, uranium has been considered in only one energy cointegration study (Mjelde and Bessler, 2009).

2

2030

2.

COMMODITY PRICE CYCLES

A common thread of the commodity price literature focuses on long-term trends in prices, following Hotelling’s seminal paper (1931) and the hypothesis proposed by Prebisch (1950) and Singer (1950). 5 However Cashin and McDermott (2002) show, relative to any trend that may exist, that commodity prices exhibit large variability. One way to decompose this volatility is to consider the cyclical behaviour of these prices. Commodities display cyclical patterns in price because short-run supply and demand are generally inelastic. Wheat farmers cannot increase output until the next season, and there are limits to increasing production from existing mines for mineral commodities. Market shocks therefore tend to cause large fluctuations in price before quantity adjusts in the medium to long run. Newbery and Stiglitz (1981, p. 152) produce a general dynamic system which experiences random shocks will experience cyclical price movements. This chapter follows the methodology of Cashin et al. (2002), who use the Bry-Boschan algorithm to investigate the cycle properties of a range of commodities. They find that slumps in commodity prices are longer than booms, and that the magnitude of a given boom or slump is generally not related to its length. 6 As far as we are aware, the uranium price has not been analysed using this method. We compare uranium to a sample of 19 other commodities. In the uranium market, the combination of highly inelastic short-term demand and large shifts in expectations may exacerbate volatility, producing relatively few booms and slumps with large price changes. We find that uranium experiences large booms and slumps, relative to other commodities. While slumps in the uranium price are the longest of all commodities, booms are much shorter than the average.

2.1

Data

The uranium price between 1974 and 2013 is from the Bureau of Resource and Energy Economics’ (BREE) Resources and Energy Statistics, with data going back to 1971 extrapolated from the annual price series from the OECD-NEA’s Forty Years of Uranium Resources, Production and Demand in Perspective. All other prices are from the World Bank’s Commodity Prices Data. All prices are deflated by the Goods Export Unit Value index for advanced economies, from the IMF’s International Financial Statistics. The real commodity prices therefore reflect their price relative to traded manufactured goods. The cycles were dated using the log of the real price multiplied by 100. However, peak and trough selection is invariant to both transformations.

2.2

Dating cycles with the Bry-Boschan algorithm

The Bry-Boschan algorithm dates peaks and troughs in a series by isolating local maxima and minima, and then applying a set of censoring rules. 7 The algorithm is based on the graphical methods used by Burns and Mitchell (1946), to date turning points in the business cycle; Bry and Boschan (1971) automated this procedure. Preliminary peaks are selected by isolating local maxima, which occur in a series 𝑝𝑝𝑡𝑡 occurs at time 𝑡𝑡 when: 𝑝𝑝𝑡𝑡−𝑘𝑘 < 𝑝𝑝𝑡𝑡 > 𝑝𝑝𝑡𝑡+𝑘𝑘

∀ 𝑘𝑘 = 1, … , 𝐾𝐾

(2.1)

where 𝐾𝐾 is five for monthly data. Preliminary troughs are identified similarly. The censoring rules then ensure that the turning points selected represent a cycle, where peaks and troughs strictly alternate. In the case of a double peak (or trough), the earlier turning point is eliminated. The phase from a trough to a peak is called a boom, and between a peak and a trough is called a slump. The algorithm forces a single phase to be at least six months long, while a full cycle must be longer than 15 months. These restrictions are in line with Roberts (2009) and Chen et al. (2012), who use the Bry-Boschan algorithm to analyse metals prices. Cashin et al. (2002) use longer restrictions of 12 (phase) and 24 months (cycle). While the shorter restrictions appear to have identified some turning points that we would not select on observation, similarly the longer restrictions miss some turning points that we would select. Given the aim of this chapter is to systematically compare uranium to other commodities, consistency is paramount. Therefore we have applied the shorter restrictions to all commodities in our sample. 5

See for example Grilli and Yang (1988) and Cuddington and Urzúa (1989). Other work includes Labys et al. (1998) and Labys et al. (2000) who applied rules similar to the Bry-Boschan algorithm, and Roberts (2009) and Chen et al. (2012) who explicitly follow this procedure. 7 We use the ‘Excel for BBQ’ program by Sam Oularis of the IMF Institute, which is based on the GAUSS version of the original BBQ program by James Engel. Both are available from the National Centre of Econometric Research at: http://www.ncer.edu.au/data/. 6

3

Finally, a threshold is applied which allows the minimum phase length to be ignored in the case of sharp movements over a short period. The threshold was set at a log change of 0.45. This was only binding for crude oil, natural gas and wheat. For crude oil, the threshold allows the abrupt fall in price in the second half of 2008 to be considered a slump, separate from the subsequent recovery (see Appendix). The length of each phase is the duration, and the price change over each phase the amplitude. We also consider the correlation between duration and amplitude, and calculate the excess index to explain the path of growth within each phase (Harding and Pagan 2002). Correlation of duration and amplitude We consider whether there is a relationship between duration and (absolute) amplitude using the Spearman rank correlation coefficient. A positive coefficient implies that longer booms (or slumps) are associated with larger price movements. Excess index If a price grows at a constant rate over a phase, the time path would form the hypotenuse of a right-angled triangle where the base is the duration and the height is amplitude. In reality the actual time path often diverges from this (Figure 2.1 is a stylised boom). The excess index quantifies this divergence by comparing the area under the actual time path to the area under the constant-growth triangle (Harding and Pagan 2002). Figure 2.1: Stylised boom and the constant-growth Figure 2.2: Stylised boom showing triangle calculation of area under actual time path

Amplitude

Actual Path

Duration

For a phase that starts at time 0 and ends at 𝑇𝑇 , the area under the actual time path is 𝐶𝐶. The difference between 𝐶𝐶 and the area of the constant-growth triangle is then averaged over all the phases for a given commodity to calculate the excess index: 𝑇𝑇 1 𝐶𝐶 = �[𝑝𝑝𝑡𝑡 − 𝑝𝑝0 ] − 𝐴𝐴; 2 𝑡𝑡=1

1 1 𝐽𝐽 𝐶𝐶𝑗𝑗 − 2 𝐴𝐴𝑗𝑗 𝐷𝐷𝑗𝑗 𝐸𝐸 = � 𝐽𝐽 𝑗𝑗=1 𝐷𝐷𝑗𝑗

(2.2)

Where 𝐴𝐴 is amplitude, 𝐷𝐷 is duration and 𝐽𝐽 is the number of booms (slumps) for the commodity. This is calculated separately for booms and slumps. The path shown in Figure 2.1 has a negative 𝐸𝐸. It is characterised by slow initial growth, which then accelerates towards the next turning point. We use a simple t-test for the null hypothesis that the mean excess index is equal to zero. The price path over a phase may not wholly be inside or outside the triangle. In this case the sign of the excess index reflects whether the area inside or outside the constant-growth triangle is larger (Chen et al., 2012, p. 59).

4

Figure 2.3: Uranium price cycles, 1971:1 to 2013:11 USD /lb 120

Uranium

80

40

0 1971

1975

1979

1983

1987

1991

1995

1999

2003

2007

2011

Notes: Booms are indicated by the shaded regions. Sources: BREE, OECD/IAEA, author’s calculations.

Table 2.1: Uranium price cycles, 1971:1 to 2013:11 Dates – 1974:03 1976:04 1982:08 1983:09 1985:04 1985:12 1991:11 1993:10 1994:10 1996:06 2000:12 2002:02 2003:03 2007:06 2008:10 2009:06 2010:04 2011:01 –

Slumps Duration Amplitude Excess Index 39 36.8 -2.44 76 124.9 -18.13 19 49.2 5.29 71 112.4 -5.26 12 18.0 0.59 54 66.1 0.53 13 10.7 0.55 16 113.5 11.29 10 24.7 5.55 34 71.7 –

Booms Duration Amplitude Excess Index 25 148.8 0.60 13 38.7 4.57 8 9.0 2.34 23 37.3 3.34 20 57.3 -4.98 14 39.3 4.40 51 237.6 -27.73 8 17.2 2.79 9 50.2 -9.02

Dates 1974:03 1976:04 1982:08 1983:09 1985:04 1985:12 1991:11 1993:10 1994:10 1996:06 2000:12 2002:02 2003:03 2007:06 2008:10 2009:06 2010:04 2011:01

Notes: Duration is in months, amplitude is in absolute log changes times 100. A positive excess index indicates a path outside the triangle approximation and shows the average excess growth in terms of log change per month times 100. Source: Author's calculations.

Figure 2.4: Millennium boom price increase, log change times 100 Growth 200

150

Beef

Aluminium

Platinum

Coffee

Gold

Maize*

Wheat*

Silver*

Palm Oil*

Natural Gas*

Nickel

Tin*

Copper

Crude Oil*

Zinc

Iron Ore

Coal*

Lead*

Phosphate Rock

50

Uranium

100

0 Notes: Growth is measured in 100 times log change. A change of 69 indicates the price doubled. Growth is measured from the first trough after 2000 to the last peak before 2009. * indicates this growth occurred in multiple boom phases. Source: Author’s calculations.

5

9 13 12 11 9 14 12 14 12 11 11 9 13 15 10 10 14 8 11 13

Number

19.0 23.7 19.9 25.7 26.3 18.1 19.3 17.4 21.3 24.0 19.4 27.2 18.3 15.5 23.1 26.3 16.5 34.4 21.9 18.2

Duration

Booms Amplitude Excess Index index sd 70.6 -2.63 10.47 67.2 -0.03 22.7 47.8 -1.54 7.91 95.7 -5.95 13.92 66.6 2.30 30.73 56.8 0.38 3.79 47.8 -5.17* 6.07 32.4 0.10 3.64 42.5 2.69 6.52 74.2 -7.15* 10.56 48.5 -3.67 9.2 48.4 -1.91 15.3 56.4 -1.49 10.84 52.1 -3.31 8.42 49.0 -2.77 7.96 74.6 4.34* 5.63 56.6 -6.01* 9.89 63.3 -7.69 12.11 51.8 -1.86 6.04 64.8 -8.48 17.02 0.78* 0.57* 0.71* 0.54* 0.86* 0.69* 0.66* 0.61* 0.22* 0.48* 0.86* 0.74* 0.91* 0.63* 0.51* 0.38* 0.26* 0.69* 0.69* 0.37*

Correlation 9 13 12 11 9 15 11 14 12 11 11 10 13 14 11 10 13 7 11 13

Number 34.3 14.0 21.2 19.3 27.7 16.2 23.5 18.0 19.9 20.8 25.1 24.2 19.7 18.7 23.6 22.8 20.2 29.9 22.8 19.9

Duration 62.8 52.1 38.6 75.5 52.6 58.9 50.3 33.1 39.0 72.5 48.5 37.9 49.5 48.4 45.8 67.8 54.1 38.5 40.3 53.0

Slumps Amplitude

Excess Index index sd -0.23 8.28 -3.82 9.81 0.02 4.32 5.44 6.91 4.95 12.32 -0.32 10.37 0.40 7.47 0.46 3.84 2.44 5.01 -0.65 8.30 -0.15 6.73 -0.96 4.88 0.49 10.16 2.74 5.79 1.15 5.79 -2.48 5.84 2.13 6.54 0.25 4.01 -1.44 3.85 1.57 6.57

0.71* -0.13* 0.05* 0.34* 0.12* 0.67* 0.66* 0.55* 0.72* 0.68* 0.05* 0.52* -0.12* 0.59* -0.37* 0.45* 0.87* 0.40* 0.56* 0.43*

Correlation

0.48 11.6 21.8 58.4 -2.49 -0.61* 11.5 22.1 51.0 0.60 -0.40* Average Notes: Duration, amplitude and excess indices are averages. Duration is in months, and amplitude in absolute log change times 100. The excess index is measured in excess growth per month, correlation is Spearman's rank correlation coefficient between duration and (absolute) amplitude. * is significant at the 5 per cent level. Source: Author's calculations.

Uranium Crude Oil Coal Natural Gas Phosphate Rock Palm Oil Maize Beef Wheat Coffee Aluminium Iron Ore Copper Lead Tin Nickel Zinc Gold Platinum Silver

Proportion of time in boom 0.33 0.65 0.46 0.55 0.46 0.53 0.45 0.47 0.50 0.51 0.41 0.53 0.46 0.45 0.49 0.51 0.45 0.54 0.47 0.46

Table 2.2: Price cycle summary statistics, all commodities, 1971:1 to 2013:11

2.3

Results

Table 2.2 outlines the statistics calculated for each commodity. Broadly speaking these results concur with previous work. The discrepancies can be attributed to different samples or different parameters in the censoring rules. For example, we calculate a higher average boom amplitude for most commodities than Cashin et al. (2002, p. 285) as we can include the millennium boom in our sample. Of all commodities uranium spends the least time in the boom phase. On average commodities are in a boom 48 per cent of the time, yet for uranium it is only 33 per cent. Over the period uranium completes relatively few full cycles; only gold completes fewer. The uranium price is characterised by two distinct booms from 1974 to 1976 and 2003 to 2007, with an extended period of depressed prices between (Figure 2.3; Table 2.1). Figure 2.5: Average boom duration (months), 1971:1 to 2013:11 Gold Iron Ore Phosphate Rock Nickel Natural Gas Coffee Crude Oil Tin Platinum Wheat Coal Aluminium Maize Uranium Copper Silver Palm Oil Beef Zinc Lead

Figure 2.6: Average boom amplitude (100 times log change), 1971:1 to 2013:11 Natural Gas Nickel Coffee Uranium Crude Oil Phosphate Rock Silver Gold Palm Oil Zinc Copper Lead Platinum Tin Aluminium Iron Ore Coal Maize Wheat Beef

34.4 27.2 26.3 26.3 25.7 24.0 23.7 23.1 21.9 21.3 19.9 19.4 19.3 19.0 18.3 18.2 18.1 17.4 16.5 15.5 0

10

20

95.7 74.6 74.2 70.6 67.2 66.6 64.8 63.3 56.8 56.6 56.4 52.1 51.8 49.0 48.5 48.4 47.8 47.8 42.5 32.4 0

30

20

40

60

80

Source: Author’s calculations.

The average boom phase for uranium is shorter than most, at 19 months (Figure 2.5). While its booms are relatively short, the average price increase is relatively large, at 103 per cent (Figure 2.6). Only natural gas, nickel and coffee have larger average boom amplitudes, but this growth occurs over much longer durations of around 25 months. However, the average for uranium is dominated by the two large booms. Uranium has the longest average slump duration of any commodity, at 34 months; the average duration for all commodities is only 22 months (Figure 2.7). In the slump phase, uranium has the fourth largest average amplitude (Figure 2.8). The four commodities with the largest amplitudes during booms also experience the biggest average slumps. This highlights a broader symmetry that exists between the amplitudes of each phase; nine of the ten commodities with above-average boom amplitudes also have above-average slump amplitudes. Gold is the exception to this trend, with relatively large booms but relatively small slumps. This arguably reflects the unique nature of gold as a financial asset and safe haven for investors. Uranium and 12 other commodities show significant, positive correlations between boom amplitude and duration. A slightly different group of 13 commodities (still including uranium) have significant, positive correlations in the slump phase. This means that the longer the phase, the larger the (absolute) change in price a commodity experiences. Notably crude oil and copper actually have a significant negative correlation in the slump phase; a shorter slump in these commodities is correlated with a bigger price fall. This result differs from Cashin et al. (2002, p. 284), who find only a few commodities with significant correlations. The discrepancy

7

could arise because we impose shorter restrictions on minimum phase and cycle length, isolating short booms and slumps with small price changes. Including the millennium boom also increases the chance of finding significant correlations between duration and amplitude. Figure 2.7: Average slump duration (months), 1971:1 to 2013:11 Uranium Gold Phosphate Rock Aluminium Iron Ore Tin Maize Nickel Platinum Coal Coffee Zinc Wheat Silver Copper Natural Gas Lead Beef Palm Oil Crude Oil

Figure 2.8: Average slump amplitude (100 times log change), 1971:1 to 2013:11 Natural Gas Coffee Nickel Uranium Palm Oil Zinc Silver Phosphate Rock Crude Oil Maize Copper Aluminium Lead Tin Platinum Wheat Coal Gold Iron Ore Beef

34.3 29.9 27.7 25.1 24.2 23.6 23.5 22.8 22.8 21.2 20.8 20.2 19.9 19.9 19.7 19.3 18.7 18.0 16.2 14.0 0

10

20

75.5 72.5 67.8 62.8 58.9 54.1 53.0 52.6 52.1 50.3 49.5 48.5 48.4 45.8 40.3 39.0 38.6 38.5 37.9 33.1 0

30

20

40

60

Source: Author’s calculations.

Uranium’s dual booms 8 We consider the two uranium price booms in turn. The 1970s boom is extensively explored in the uranium literature. 9 The threefold increase in price was driven by a range of factors: unrealistic demand expectations; the US government monopoly in enrichment services (natural uranium must be enriched before it is used to generate power); the creation of a cartel; and supply delays from major producers. The end of the boom coincided with the opening of European enrichment facilities, the easing of the embargo on foreign uranium use in US reactors and the revelation of the cartel (and subsequent legal battle). The slump that followed lasted more than seven years and reversed much the price increase. The Three Mile Island disaster during this slump led to a reduction in the number of proposed reactors, dampening expectations of future demand. The boom phase for uranium from 2003 to 2007 coincides with the millennium commodities boom. This is usually associated with rapid increases in coal and iron ore prices driven by demand from developing markets. Indeed the prices of coal and iron ore both quadrupled. Though price growth was substantial for all commodities, uranium experienced the greatest growth over the millennium boom of any commodity in our sample (Figure 2.4). The price increased tenfold from US$12.46 in 2003 to US$125.22 at its peak in June 2007. The Bry-Boschan algorithm shows that the period from 2000 to 2008 was not a continuous increase in price for most commodities. Generally any given commodity was only in a boom phase for a portion of the period, or experienced two or three separate booms. Iron ore was an exception: it was in a boom for eight years from the end of 1999 to November 2007. The reason for uranium’s millennium boom is less clear than for other commodities. Expectations of future growth in demand for nuclear capacity in China, India and Russia helped fuelled the price increase. This was 8 This cycle analysis sheds little light on the effect of the three biggest nuclear disasters on the uranium price. The uranium price was in a slump phase before each disaster occurred (Three Mile Island, March 1979; Chernobyl, April 1986; Fukushima-Daiichi, March 2011). These disasters may have increased the length of the slump phase they occurred in, but testing the counterfactual is difficult. 9 Taylor and Yokell (1979), Radetzki (1981), Owen (1985).

8

perhaps buoyed by a renewed sense that nuclear power could provide baseload energy cheaply and with low carbon emissions (OECD/IAEA, 2014). At the same time there were production delays in both Canada and Australia, and continued uncertainty about the level of inventories and secondary supply (OECD/IAEA, 2006). This points to large movements in the price being driven by the highly inelastic demand for uranium and its low share of total generation costs. The excess index and the price growth path The excess index quantifies the nature of growth within each phase. For uranium, the excess index is negative in both booms and slumps, producing a time path inside the constant-growth triangle (Figure 2.1). This implies that the change in the uranium price is largest on either side of the peak; growth accelerates in the latter part of the boom, and the price then falls quickly before growth slows as it approaches the trough. Most commodities have negative excess indices in the boom phase (Table 2.2). Commodities are more evenly split in the slump phase. However, average index values disguise significant variation between phases for a given commodity. Tests indicate that the average excess index is not significantly different from zero for most commodities, including uranium. This limits the conclusions that can be drawn about the growth path of commodities. Ingram (2015) applies a Markov-Switching State Space model in conjunction with the Bry-Boschan algorithm to investigate this issue; he finds that most of the price change occurs at the end of a phase. It is more appropriate to discuss the excess index for a given phase. In the 1970s boom, the excess index for uranium was around zero, implying a constant rate of growth. The excess index was strongly negative in the millennium boom; the fastest growth in the uranium price occurred at the end of the period (Figure 2.3).

2.4

Conclusion

While it is clear that the uranium price exhibits the cyclical behaviour common to commodities, the cycles on average appear to be more extreme. The uranium price exhibits relatively long periods in a slump, punctuated by short booms. The amplitude of the price movements within each phase are large when compared to the sample of commodities. These price dynamics could reflect the combination of exuberant expectations, the insensitivity of demand to the uranium price and short-term supply inelasticity. Of course, with this methodology we cannot directly ascribe the characteristics of uranium’s price cycles to the nature of the market. However, this provides evidence that the structural rigidities in the market affect the behaviour of the uranium price.

3.

INTEGRATION OF REGIONAL URANIUM PRICES

The title of this paper implies that a single global market for uranium exists. Here we consider a market to be the “area in which price is determined” (Stigler and Sherwin, 1985, p. 555). Uranium is a relatively homogenous product with a high value-weight ratio, so prices should theoretically equalise between regions. However, substantial structural barriers exist: there is no spot market to arbitrage away regional discrepancies; inactive futures markets discourage speculators; and long-term contracts limit the ability for producers and consumers to adjust prices, especially in the short run. Bilateral safeguards agreements may also limit competition, but Australia, Canada and Euratom have signed agreements with most, if not all, major consumers and producers (DFAT, 2014, DFATD, 2014, European Commission, 2014). In lieu of a spot market, participants might instead rely on price indicators produced by two private firms, TradeTech and UxConsulting. TradeTech (2014) claims that its price indicators are “widely referenced as settlement prices for long-term contract deliveries.” The question still remains as to what role the price indicators actually play in the market. We will call the average of these price indicators the ‘spot price’. Theoretically the price indicators could substitute for a liquid spot market, acting as a signal for the rest of the market. Producers and consumers could use them as the basis for decisions if they were believed to be a credible, unbiased reference point. In this chapter we consider whether the spot price is an important reference price, by testing the degree of integration between it and three regional prices (Figure 3.1). For other fuels there is evidence that regional markets are integrated. Bachmeier and Griffin (2006) find strong evidence for a world market for crude oil where equilibrium is restored relatively quickly. Transport costs are more likely to segment natural gas and coal markets. Bachmeier and Griffin (2006) also suggest that five coal prices in the United States are cointegrated, but the relationship is much weaker. Regulatory changes at the end of the 1970s allowed a national spot market

9

for natural gas to develop in the United States (Doane and Spulber, 1994). This is reaffirmed in other work (Serletis and Rangel-Ruiz, 2004, Gebre-Mariam, 2011). In the uranium market, we find that the spot price is cointegrated with each of the three regional prices, and that it is weakly exogenous in each relationship. Therefore only the regional prices adjust to correct deviations from the long-run equilibrium. The price indicators reflect the conditions in the uranium market, and are a viable substitute for price discovery in lieu of a formal exchange. Figure 3.1: World uranium prices USD/kg 240 200 Spot

160

Australian Export 120 80 US Domestic

40

European Spot 0 1970

1974

1978

1982

1986

1990

1994

1998

2002

2006

2010

Notes: Prices are per kilogram of pure uranium. Sources: BREE, EIA, ESA, OECD/IAEA.

3.1

Data

Regional uranium prices are only available annually. Given the market is dominated by multi-year contracts rather than spot trading, the risk of losing important dynamics by averaging over a year is low. However the econometric disadvantages of a small dataset remain. The spot price is an average of the price indicators. The Australian price is the average export price. The European spot price is the average price paid by utilities in the European Union for deliveries under short-term contracts. The US domestic price is the average price for domestic purchases. Sources for these data are in the Appendix. All prices are quoted in US dollars per kilogram of pure uranium, with volume conversions based on TradeTech ratios. 10 Where necessary, prices were converted to US dollars using exchange rates quoted by the respective source. Prices have been deflated by the Manufacturing Unit Value index from the World Bank. All models are estimated using the log of real prices. The results of Augmented Dickey-Fuller (ADF) tests are presented in the Appendix. We do not reject the null of a unit root process for any series. The first difference is clearly stationary and therefore each series is integrated of order one.

3.2

A single market for uranium

We consider prices to be linked if a stationary relationship exists in the long run, and the prices return to equilibrium following an exogenous shock. We test for cointegration, following Johansen (1988). A group of non-stationary variables are cointegrated if there exists a linear combination that is stationary (Engle and Granger, 1987, p. 253). Tests for cointegration were conducted both with and without an intercept term (Table 3.1). We can reject the null hypothesis of no cointegration at the five per cent level for the spot price with the Australian and US prices, but not the European price. However, when the intercept is excluded all three regional prices are cointegrated with the spot price. Excluding the intercept is possibly an overly restrictive assumption, as it implies there is no gap between the prices in the long run.

10

Available at http://www.uranium.info/unit_conversion_table.php.

10

Table 3.1: Bivariate cointegration tests with the spot price and regional prices Australian Export 1983 – 2012 r Trace Intercept in cointegrating equation 0 31.37*** (0.001) 1 4.08 (0.401) No intercept 0 25.46*** (0.000) 1 0.00 (0.994)

European Spot 1980 – 2013

US Domestic 1981 – 2011

Max

Trace

Max

Trace

Max

27.30*** (0.001) 4.08 (0.401)

15.88 (0.180) 2.50 (0.678)

13.38 (0.119) 2.50 (0.678)

28.72*** (0.002) 2.29 (0.719)

26.43*** (0.001) 2.29 (0.719)

25.46*** (0.000) 0.00 (0.994)

13.30** (0.034) 0.04 (0.870)

13.26** (0.022) 0.04 (0.870)

26.36*** (0.000) 0.02 (0.911)

26.34*** (0.000) 0.02 (0.911)

Notes: Tests were conducted over the longest sample available. The null hypothesis of the trace test is the number of cointegrating relationships is equal to or less than r, against an alternative that the number is greater than r. The null hypothesis of the max test is the number of cointegrating relationships is equal to r, against the alternative that the number is equal to r+1. P-values in parentheses. *, **, and *** indicate rejection of the null at the 10 per cent, 5 per cent and 1 per cent level. Source: Author's calculations.

There is sufficient evidence of cointegration to suggest integration between the regional prices and the spot price. Cointegrated variables can be represented in an error-correction form (Engle and Granger, 1987). We therefore estimate a Vector Error-Correction Model (VECM) for each pair, of the form (Table 3.2): Reg Reg Spot Spot Δ𝑝𝑝𝑡𝑡Spot = 𝛼𝛼1 �𝛽𝛽0 + 𝑝𝑝𝑡𝑡−1 + 𝛽𝛽1 𝑝𝑝𝑡𝑡−1 � + 𝜙𝜙11 Δ𝑝𝑝𝑡𝑡−1 + 𝜙𝜙12 Δ𝑝𝑝𝑡𝑡−1

Reg Reg Spot Spot Δ𝑝𝑝𝑡𝑡Reg = 𝛼𝛼2 �𝛽𝛽0 + 𝑝𝑝𝑡𝑡−1 + 𝛽𝛽1 𝑝𝑝𝑡𝑡−1 � + 𝜙𝜙21 Δ𝑝𝑝𝑡𝑡−1 + 𝜙𝜙22 Δ𝑝𝑝𝑡𝑡−1

(3.1)

Where 𝑝𝑝𝑡𝑡Spot is the spot price and 𝑝𝑝𝑡𝑡Reg is the regional price. Likelihood ratio tests indicate that the intercept is not significantly different from zero in any model (Table 3.2). This implies that in the long run the prices are proportional to each other: 𝑝𝑝𝑡𝑡Spot + 𝛽𝛽1 𝑝𝑝𝑡𝑡Reg = 0 𝑝𝑝𝑡𝑡Spot = −𝛽𝛽1 𝑝𝑝𝑡𝑡Reg

(3.2)

And if 𝛽𝛽1 is −1 the two prices will be equal in the long run. At the five per cent level, we do not reject the null hypothesis that the restriction (𝛽𝛽1 = −1) is not binding. Therefore each regional price is equal to the spot price at equilibrium: 𝑝𝑝𝑡𝑡Spot = 𝑝𝑝𝑡𝑡Reg

(3.3)

This implies there is no multiplicative regional premium or discount, which could have been caused by transport costs or structural barriers. This suggests a relatively strong level of integration. It provides evidence of a single world price for uranium that is not distorted by regional differences or transport costs. Tests are also conducted on the speed of adjustment coefficients. The spot price is weakly exogenous because we cannot reject the null hypothesis that the restriction (𝛼𝛼1 = 0) is not binding. Therefore the spot price does not adjust to restore deviations from the long-run equilibrium. The speed of adjustment coefficient on the regional price equation is statistically different from zero in all three models. When the spot price is higher than the regional price the error-correction term is positive; it is the regional price that adjusts, rising to correct the deviation from equilibrium. The three models are re-estimated by imposing the non-binding restrictions on the intercept, the cointegrating coefficient and the speed of adjustment for the spot price equation (Table 3.3). The null hypothesis that the restrictions are jointly not binding cannot be rejected at the five per cent level in any of the three models.

11

Table 3.3: Bivariate VECMs with the spot price, with intercept 𝛽𝛽0

H0: 𝛽𝛽0 = 0

𝛽𝛽1

H0: 𝛽𝛽1 = −1

𝛼𝛼1

H0: 𝛼𝛼1 = 0

𝛼𝛼2

H0: 𝛼𝛼2 = 0

𝜙𝜙11 ϕ12 𝜙𝜙21 𝜙𝜙22 2 𝑅𝑅Spot 2 𝑅𝑅Reg

Australian Export Coefficient Test 1.150 1.84 (0.175) -1.254 3.08 (0.215) -0.065 1.87 (0.393) 0.287 23.64*** (0.000) Coefficient Standard Error 0.354 [0.231] -0.334 [0.309] 0.019 [0.100] -0.573 [0.133] 0.15 0.67

European Spot Coefficient Test -0.075 0.12 (0.730) -1.003 5.59* (0.061) 0.436 2.02 (0.365) 0.799 12.42*** (0.002) Coefficient Standard Error 0.210 [0.365] -0.299 [0.231] 0.322 [0.220] -0.271 [0.139] 0.21 0.71

US Domestic Coefficient Test 0.099 0.08 (0.771) -1.028 0.26 (0.877) 0.206 1.08 (0.583) 0.477 22.79*** (0.000) Coefficient Standard Error 0.320 [0.251] -0.599 [0.377] 0.027 [0.105] -0.441 [0.158] 0.17 0.66

Notes: Coefficients refer to those in (3.1). Cointegrating coefficients are normalised such that the coefficient on the spot price equals one. The test on the constant follows a 𝜒𝜒2 distribution with one degree of freedom. All further tests have two degrees of freedom, as the 𝛽𝛽0 = 0 restriction is imposed jointly with the null hypotheses listed. P-values in parentheses and standard errors in square brackets. VECM estimated with an intercept in the cointegrating equation only. *, **, and *** indicate rejection of the null at the 10 per cent, 5 per cent and 1 per cent level. Source: Author's calculations.

Together these results are striking. In the long run, the three regional uranium prices are equal to the spot price. The world price for uranium is discovered by the spot price. This is then transmitted through the error-correction mechanism to the regional prices which adjust to restore the long-run equilibrium. The Australian export price has the slowest rate of adjustment, with a half-life of 2.8 years, while the European and US prices have a shorter half-life of around 1.6 years. 11 This suggests a long period of adjustment toward equilibrium. Given the structural rigidities this market faces, the adjustment is unlikely to be fast. However, the prices still clearly move together. The market is not as tightly integrated as other fuels with greater liquidity, but this result shows that relatively strong integration that can be achieved despite of the barriers to arbitrage. The nature of trade covered by these regional prices can explain the variation in speeds of adjustment. Euratom is a party to all transactions with European utility companies, and actively pursues a policy of supply security Table 3.2: Bivariate VECMs re-estimated with non-binding restrictions imposed 𝛽𝛽0 𝛽𝛽1 𝛼𝛼1 𝛼𝛼2 𝜙𝜙11 ϕ12 𝜙𝜙21 𝜙𝜙22 2 𝑅𝑅Spot 2 𝑅𝑅Reg

Australian Export 𝜒𝜒2 (3) Test Coefficient 0 3.10 (0.376) -1 0 0.252 Coefficient Standard Error 0.407 [0.221] -0.303 [0.317] 0.104 [0.103] -0.612 [0.148] 0.14 0.61

European Spot 𝜒𝜒2 (3) Test Coefficient 0 6.55* (0.088) -1 0 0.410 Coefficient Standard Error 0.331 [0.328] -0.323 [0.231] 0.543 [0.211] -0.315 [0.149] 0.20 0.66

US Domestic 𝜒𝜒2 (3) Test Coefficient 0 1.27 (0.736) -1 0 0.423 Coefficient Standard Error 0.332 [0.249] -0.592 [0.380] 0.041 [0.105] -0.448 [0.160] 0.16 0.66

Notes: Coefficients refer to those in (3.1). Restrictions imposed are those found to be non-binding in Table 3.2. Cointegrating coefficients are normalised such that the coefficient on the spot price equals one. P-values in parentheses and standard errors in square brackets. VECM estimated with an intercept in the cointegrating equation only. *, **, and *** indicate rejection of the null at the 10 per cent, 5 per cent and 1 per cent level. Source: Author's calculations.

11

Half-life (−ln 0.5 /𝛼𝛼) is the time to correct 50 per cent of the deviation from equilibrium.

12

by procuring uranium from a diverse range of sources (ESA, 2014, p. 29). Utilities in the United States now get most of their uranium from imported sources, introducing foreign price competition (EIA, 2014). Thus they are exposed to a larger number of transactions and adjust more quickly. However the Australian price is based on the production of only a few mines.

3.3

Conclusion

These results suggest it is appropriate to talk about a single global uranium market. While it faces significant barriers to arbitrage, the price indicators bridge the gaps between potentially segmented markets. The regional prices react to changes in the price indicators, and so they act as a signal for regional prices. This adds weight to the findings in the previous chapter, because it is clear that conclusions about the uranium market can be drawn from movements in this spot price. The price indicators are clearly influential in the market, even if not all contracts are tied to them directly. The small sample size somewhat limits these results. The coefficients may not be stable, and the results are not as robust as they otherwise would be with a larger sample size. Nevertheless these results provide new econometric evidence about the nature of the uranium market.

4.

COPRODUCTS AND ENERGY SUBSTITUTES

The previous chapter provided evidence of a single market for uranium. But the true market for uranium might be broader; the price of uranium may be determined in concert with that of other commodities. In this chapter, we focus on pairwise relationships between uranium and both its energy substitutes and coproducts. We consider three energy substitutes for uranium: coal, crude oil and natural gas (Figure 4.1). One could argue for a single market for ‘energy’, but direct competition between the fuels is limited. Therefore we focus only on bilateral relationships. The three coproducts considered are gold, phosphate rock and copper. Uranium is mined in some cases from the same deposit as these products; the joint supply decision could lead to price changes in uranium’s coproducts influencing the uranium market. The relationship between the uranium price and the price of energy substitutes and coproducts has previously been explored in supply-demand models. Kahouli (2011) and Amavilah (1994) found a contemporaneous positive relationship between the uranium price and the coal price. This could be explained by common shocks in demand for electricity, such as growth in industrial production. Kahouli (2011) also finds that the gold price has a significant positive effect on uranium supply, suggesting that an increase in the price of gold will stimulate production of uranium mined alongside gold. Figure 4.1: Real fuel commodity prices, 1979-2013 USD

USD

120

12 Natural Gas (RHS)

Coal (LHS)

8

80 Crude Oil (LHS)

4

40

Uranium (LHS) 0 1979

0 1983

1987

1991

1995

1999

2003

2007

2011

Notes: The coal price is per metric tonne, crude oil per barrel, natural gas per million British thermal units, and uranium per pound of natural uranium. Sources: BREE, OECD/IAEA, World Bank.

13

However, we find there is little evidence of integration between uranium and other fuels. In pairwise tests with uranium, crude oil is the only fuel that is integrated. This may merely reflect the importance of the world oil price in energy trade, and the economy more generally. We also find no evidence of integration between the uranium price and the prices of gold, phosphate rock, or copper.

4.1

Substitutes in energy production

Crude oil, natural gas, coal and uranium together provide 87 per cent of world energy supply, but the four fuels are not equally represented in different uses (IEA, 2013). Crude oil and petroleum products dominate transport, while more than half of the world’s electricity is generated from coal (Figure 4.2). The ability to substitute between fuels is restricted even for a given use. Only 15 per cent of natural gas capacity and a quarter of coal capacity can be switched to other fuels in the US manufacturing sector (EIA, 2013). In electricity generation, few power plants have the ability to swap between fuels. However, direct fuel substitution may occur by selecting which power plants to operate (if spare capacity exists) and in the long run through the choice of which power plants to build or decommission. Figure 4.2: World energy use by fuel type Transport Crude Oil

Industry

Natural Gas Coal

Electricity

Uranium Other Heat* 0.0

0.2

0.4

0.6

0.8

share

Notes: 2011 data for all except heat, which is from 2009. * Heat is not a mutually exclusive category, calculated separately by the IEA. Crude oil includes all petroleum products. Other includes renewable energy, and electricity for other uses. Source: IEA.

While switching capacity is small, the decision to switch could have a significant impact on fuel prices because it is made at the margin. This may be helped by rules of thumb based on relative energy content, which are common in the pricing of crude oil and natural gas (Brown and Yücel, 2007, p. 4). Agents in the market could tie prices together by relying on these rules of thumb. In the past most Liquefied Natural Gas (LNG) contracts have been indexed to the oil price (Foss, 2005), though greater flexibility in contract arrangements today weakens this link (Hartley, 2013). However, the growth of a global market for natural gas, facilitated by LNG’s low transport costs, may allow for more integration with the global crude oil market. The ability to substitute between uranium and other fuels is clearly more limited; uranium is only used to generate electricity and a nuclear power plant cannot switch fuels. Beck and Solow (1994) argue that variables related to electricity demand, including energy substitutes, will not have a significant effect on the uranium market because nuclear power’s marginal cost is low. Thus output from a given reactor is maximised in the short run (as average cost is decreasing in output), and other fuels are varied to account for peak demand. This reduces the likelihood of uranium being integrated with other fuel markets. Past work on energy market integration has focused on the relationship between natural gas and oil prices. Villar and Joutz (2006) find a cointegrating relationship, with a trend, in monthly data. Brown and Yücel (2007) use weekly prices and find a stable relationship. They also test the impact of other factors such as weather on the relationship. Hartley et al. (2008) suggest that the trend term is capturing technological improvements, and replace it with an explicit variable in their model of natural gas, crude oil, and residual fuel oil. Bachmeier and Griffin (2006) test crude oil, natural gas and coal, and show that the three markets are only weakly integrated. 12

12 Other work includes Serletis and Herbert (1999), Serletis and Rangel-Ruiz (2004), Asche et al. (2006), Panagiotidis and Rutledge (2007).

14

Uranium was first considered in a cointegration setting by Mjelde and Bessler (2009) alongside crude oil, natural gas, coal and four regional electricity markets in the United States. Using weekly data between 2001 and 2008, they find evidence of cointegration between the eight prices. All energy prices except uranium are weakly exogenous, which implies that price discovery occurs in the markets for crude oil, natural gas, and coal while equilibrium is restored by the uranium and electricity prices.

4.2

Coproducts

The proportion of uranium produced as a coproduct alongside gold, phosphate and copper has varied over time, from around 20 per cent in the 1990s to only 6.5 per cent today (Owen, 1992, OECD/IAEA, 2014). Coproducts benefit from reduced exploration expenditure and shared extraction costs. In South Africa uranium is almost exclusively produced from gold mines, with production highest between 1975 and 1985. Production slowed considerably due to low uranium prices in the 1980s (OECD-NEA, 2006, p. 107). In 1990, 28 per cent of uranium produced in the US was a coproduct of phosphate, though this production declined at the end of the 1990s (Owen, 1992, WNA, 2014). The Olympic Dam mine in South Australia, the largest single uranium deposit in the world, also produces copper, gold, and silver. Uranium is a small proportion of total revenue (BHP Billiton, 2013). Theoretically speaking, coproduct price fluctuations could affect the price of uranium by influencing supply from coproduct mines. Kahouli (2011, p. 367) finds that the gold price has a significant positive effect on the uranium supply. This would put downward pressure on the price.

4.3

Results

The data are the same as used in chapter 2, and ADF tests showing that all variables are non-stationary are in the Appendix. We first consider the energy market, then the relationship between uranium and its coproducts. Energy substitutes Cointegration tests indicate no evidence of uranium being integrated with coal or natural gas markets (Table 4.1). However, the null hypothesis of no cointegration is rejected in both tests for crude oil. Table 4.1: Bivariate cointegration tests with uranium r 0 1

Crude Oil Trace Max 15.75** 14.61** (0.046) (0.044) 1.15 1.15 (0.284) (0.284)

Natural Gas Trace Max 7.15 5.51 (0.560) (0.676) 1.64 1.64 (0.200) (0.200)

Coal Trace 12.91 (0.118) 1.62 (0.203)

Max 11.28 (0.141) 1.62 (0.203)

Notes: See Table 3.1 for hypotheses. P-value in parentheses. Schwarz Information Criterion selects two lags for an unrestricted VAR. Cointegration tested with an intercept. *, **, and *** indicate rejection of the null at the 10 per cent, 5 per cent and 1 per cent level. Source: Author's calculations.

A VECM is estimated with uranium and crude oil, taking the form: U oil oil oil U U Δ𝑝𝑝𝑡𝑡U = 𝛼𝛼1 (𝛽𝛽0 + 𝑝𝑝𝑡𝑡−1 + 𝛽𝛽1 𝑝𝑝𝑡𝑡−1 ) + 𝜙𝜙11 Δ𝑝𝑝𝑡𝑡−1 + 𝜙𝜙12 Δ𝑝𝑝𝑡𝑡−2 + 𝜙𝜙13 Δ𝑝𝑝𝑡𝑡−1 + 𝜙𝜙14 Δ𝑝𝑝𝑡𝑡−2 U oil oil oil U U Δ𝑝𝑝𝑡𝑡oil = 𝛼𝛼2 (𝛽𝛽0 + 𝑝𝑝𝑡𝑡−1 + 𝛽𝛽1 𝑝𝑝𝑡𝑡−1 ) + 𝜙𝜙21 Δ𝑝𝑝𝑡𝑡−1 + 𝜙𝜙22 Δ𝑝𝑝𝑡𝑡−1 + 𝜙𝜙23 Δ𝑝𝑝𝑡𝑡−1 + 𝜙𝜙24 Δ𝑝𝑝𝑡𝑡−2

(4.1)

The uranium price coefficient is normalised to one in the cointegrating relationship: 13 U oil 0.774 + 𝑝𝑝𝑡𝑡−1 − 1.059𝑝𝑝𝑡𝑡−1 (4.2) (0.644) − − − (0.029) The rest of the results are presented in Table 4.2. Likelihood ratio tests on the speed of adjustment coefficient indicate that neither crude oil nor uranium are weakly exogenous in this model. Both prices adjust to restore deviations from equilibrium, but the small coefficients on the speed of adjustment indicate this process is very slow. Though they are cointegrated, the integration between the two markets is weak. The low 𝑅𝑅2 values reinforce this argument.

13

Standard errors in parentheses.

15

This weak integration makes sense in the context of the mechanism through which the crude oil price could affect the uranium price. There is very little opportunity for substitution between uranium and crude oil (oil provides less than seven per cent of electricity production). The relationship we find is most likely driven by crude oil’s importance to the macroeconomy. The literature on the impact of oil shocks on both equity markets and the economy is large (see for example Hamilton, 2003). The crude oil price could be acting as a proxy for sentiment in energy markets, or economic activity. Table 4.2: Bivariate Vector Error-Correction Model, uranium and crude oil. 𝛼𝛼𝑖𝑖

H0: 𝛼𝛼𝑖𝑖 = 0

𝜙𝜙𝑖𝑖1 ϕ𝑖𝑖2 𝜙𝜙𝑖𝑖3 𝜙𝜙𝑖𝑖4 𝑅𝑅2

Uranium (𝑖𝑖 = 1) Coefficient Test 6.091** -0.017 (0.014) Standard Error -0.064 [0.031] -0.012 [0.031] 0.323 [0.049] 0.015 [0.049] 0.127

Crude Oil (𝑖𝑖 = 2) Coefficient Test 7.982*** 0.030 (0.005) Standard Error 0.204 [0.048] -0.036 [0.049] -0.042 [0.076] 0.189 [0.076] 0.075

Notes: Coefficients refer to those in (4.1). The null hypothesis for the speed of adjustment test is that it is not significantly different from zero. The test statistic follows a 𝜒𝜒 2 distribution with one degree of freedom. P-values in parentheses and standard errors in square brackets. VECM estimated with an intercept in the cointegrating equation only. *, **, and *** indicate rejection of the null at the 10 per cent, 5 per cent and 1 per cent level. Source: Author's calculations.

Past supply-demand models – which found no relationship between crude oil and uranium – only allowed the current crude oil price to affect the current uranium price, and not the other way around. But in an errorcorrection framework, we find there is a tendency for crude oil and uranium to move (albeit slowly) towards a long-term equilibrium. Both the uranium and crude oil prices adjust towards this equilibrium. When considering non-stationary variables, cointegration provides a clearer understanding of their relationship. We may have found little evidence of integration between uranium and other fuels if the transmission mechanism is not included in our models. Mjelde and Bessler (2009) find that regional electricity prices are important in restoring equilibrium in their model of the four fuels. Hartley et al. (2008) find that residual fuel oil links the crude oil and natural gas markets. Further work on the uranium market could investigate whether other prices in the energy market link the uranium price to other fuels. However, this paper shows there is little convincing evidence of strong integration between the prices of uranium and other fuels. Coproducts There is no evidence that a long-run relationship exists between uranium and any of its coproducts (Table 4.3). This is likely because coproduct production is limited to a few geographical locations, and has varied in importance over time. However, we also test the period in which production was greatest for each coproduct source.14 The result is the same. Table 4.3: Coproduct cointegration tests Gold r 0 1

Trace 14.87 (0.234) 2.41 (0.695)

Max 12.45 (0.161) 2.41 (0.695)

Phosphate Rock Trace Max 11.74 10.44 (0.473) (0.296) 1.29 1.29 (0.909) (0.909)

Copper Trace 7.56 (0.859) 1.57 (0.860)

Max 5.98 (0.790) 1.57 (0.860)

Notes: See Table 3.1 for hypotheses. P-value in parentheses. Schwarz Information Criterion selects two lags for an unrestricted VAR. Cointegration tested with an intercept. . *, **, and *** indicate rejection of the null at the 10 per cent, 5 per cent and 1 per cent level. Source: Author's calculations.

Instead of linking the prices together, the presence of uranium in a coproduct deposit could act more like a real option. An increase in the coproduct price may result in more uranium reserves being discovered, but the miner is not obligated to exploit the uranium ore even if the coproduct is extracted. In South Africa, uranium ore in 14

1975-1985 for gold, 1990-1998 for phosphate, and 1990-2013 for copper.

16

gold deposits was initially discarded before being reprocessed later (Owen, 1992). Although the break-even cost for coproduct-produced uranium is perhaps lower as costs are shared between products, the decision to process the ore is still clearly driven by the uranium price.

4.4

Conclusion

The uranium price does not appear to be integrated with other related commodity prices, with the exception of the weak relationship with the crude oil price. Even this relationship may only exist because crude oil is acting as a proxy for energy demand or conditions in the macroeonomy. These results differ from the uranium supply-demand literature. In contemporaneous regressions the price of coal was found to be a significant determinant of the uranium price, but the crude oil price was not (Kahouli, 2011). A contemporaneous relationship is different to cointegration. The contemporaneous coal price may have a significant impact on the uranium price because they are related through common demand factors in electricity generation, such as growth in industrial production. However, in a cointegration framework they are not integrated. Uranium-specific factors therefore must be the most important forces determining the uranium price. The lack of integration with other energy markets may imply that uranium is different to other fuels. But the links between the other fuels are not overly strong and so it would be a surprising result if we did find high levels of integration. Given the small and declining importance of coproduct-mined uranium, the lack of integration with coproduct prices is also expected.

5.

SUMMARY

This paper has provided a novel understanding of the global uranium market, by analysing the cyclical behaviour of the uranium price and the level of market integration. We have established that uranium is broadly similar to other commodities. But the uranium market also displays some unique characteristics, which could be caused by the idiosyncrasies of nuclear power and the uranium trade. Uranium exhibits the cyclical behaviour observed in all commodities, although its booms and slumps are substantially larger than the average. Over the last forty years, uranium has been in a slump two-thirds of the time; other commodities spend roughly equal amounts of time in each phase. This is reflected in the relatively short booms, but relatively long slumps of the uranium price. This behaviour may be driven by the insensitivity of demand to price fluctuations, as well as the important role expectations play given the long lead times on nuclear power plants and uncertainty around other sources of supply. Though the nature of the uranium market produces arguably more extreme cycles than other commodities, the uranium market is surprisingly integrated. Uranium is traded in a single integrated market. Regional uranium prices adjust to correct deviations from the privately calculated price indicators. Though this adjustment process may be somewhat slow, this is probably because long-term contracts dominate. This paper shows that the price indicators are important reference prices for the uranium market, and play a leading role in determining prices around the world. The uranium market faces potentially large barriers: the relatively small volumes traded; the presence of strict safeguards; and the lack of a formal spot market. Despite these issues, the market is integrated. The price indicators can be considered the world price of uranium. Little clear evidence of a single market for energy was found, and the uranium market is not linked with that of natural gas or coal. Uranium is integrated with crude oil, although the relationship is weak and the two prices do not respond quickly to shocks. This does not corroborate the past empirical work on uranium supply and demand. However, it adds weight to the argument that the uranium price is not heavily influenced by substitution effects in the energy market. We also find no relationship between uranium and its coproducts. Further work on the uranium market falls into two areas. Firstly, empirical questions remain in determining what factors in the uranium market create the behaviour we observe. Anecdotally the issues we have discussed could explain the empirical regularities explored in this paper. However the relative significance of each factor is unknown. Secondly, the analysis of uranium in an energy integration context can be extended in various ways, including testing for potential links between the uranium price and electricity prices. Going forward, the global uranium market faces an uncertain future. While strong growth in nuclear capacity is expected in China, India and Russia, this will be partially offset by the gradual phase out occurring in parts of Europe. Of course, history suggests expectations of uranium demand be treated with caution.

17

APPENDIX Table A.1: Commodities used in price cycles analysis Energy Uranium Coal Crude Oil Natural Gas

Fertilisers Phosphate Rock

Renewables Beef Coffee Maize Palm Oil Wheat

Industrial Metals Aluminium Copper Iron Ore Lead Nickel Tin Zinc

Precious Metals Gold Platinum Silver

Notes: All prices are per metric tonne, except uranium (per pound), crude oil (per barrel), natural gas (per million British thermal units), beef and coffee (per kg), and the precious metals (per troy ounce). WTI crude oil price used for commodity price cycles analysis, while the Brent price is used for cointegration analysis.

Table A.2: Price series for Chapter 3 Series

Period

Sources

Spot Price Aus Export Price

1974 - 2013 1983 - 2012

BREE (2014)

European Spot Price

1980 - 2013

ESA (2014)

US Domestic Price

1981 - 2011

EIA (2012)

OECD-NEA (2006), OECD/IAEA (2010, 2012, 2014)

Figure A.1: Crude oil – the impact of the threshold restriction Threshold USD/ bbl 90 60 30 0 No Threshold USD/ bbl 90 60 30 0 1971

1975

1979

1983

1987

1991

1995

Notes: expansions are indicated by the shaded regions. Sources: World Bank, author’s calculations.

18

1999

2003

2007

2011

Table A.3: Augmented Dickey Fuller tests No intercept or trend Statistic Probability Chapter 3 (annual) Spot Australian Export EU Spot US Domestic Chapter 4 (monthly) Crude Oil Natural Gas Coal Uranium Gold Phosphate Rock Copper

Intercept Statistic Probability

Intercept and trend Statistic Probability

First differences Statistic Probability

-0.073 0.289 -0.454 -0.125

0.65 0.76 0.51 0.63

-2.101 -1.359 -1.412 -0.923

0.25 0.59 0.56 0.77

-1.932 -0.764 -1.702 -0.376

0.62 0.96 0.73 0.98

-4.40 -1.90 -4.38 -4.01

0.00*** 0.05** 0.00*** 0.00***

0.592 -1.021 0.207 -0.206 1.545 0.072 0.080

0.84 0.28 0.75 0.61 0.97 0.71 0.71

-2.177 -2.461 -1.457 -1.257 -1.809 -2.213 -2.190

0.22 0.13 0.55 0.65 0.38 0.20 0.21

-2.508 -2.859 -1.980 -1.218 -1.938 -2.333 -2.630

0.32 0.18 0.61 0.91 0.63 0.41 0.27

-19.480 -20.127 -16.897 -15.758 16.965 -19.393 -15.965

0.00*** 0.00*** 0.00*** 0.00*** 0.00*** 0.00*** 0.00***

Notes: All variables in logs. The null hypothesis of the first test is 𝛾𝛾 = 0 in Δ𝑝𝑝𝑡𝑡 = 𝛾𝛾𝑝𝑝𝑡𝑡−1 + 𝜀𝜀𝑡𝑡 . The second test adds a constant term 𝑎𝑎0 , and the third test additionally includes a trend term 𝑎𝑎2 𝑡𝑡. The final column recomputes the test using first differences (with no intercept or trend term included). Lags were selected using the Schwarz Information Criterion, such that the error term follows a white noise process. *, **, and *** indicate rejection of the null at the 10 per cent, 5 per cent and 1 per cent level. Source: Author's calculations.

REFERENCES Ahmed, S. B. (1979) Nuclear fuel and energy policy. Massachusetts: Lexington Books. Amavilah, V. H. S. (1994) 'The influence of oil and coal prices on the world uranium demand', OPEC Review, 18(4), pp. 489-508. Amavilah, V. H. S. (1995) 'The capitalist world aggregate supply and demand model for natural uranium', Energy Economics, 17(3), pp. 211-220. Asche, F., Osmundsen, P. and Sandsmark, M. (2006) 'The UK market for natural gas, oil and electricity: are prices decoupled', The Energy Journal, 27(2), pp. 27-40. Bachmeier, L. J. and Griffin, J. M. (2006) 'Testing for market integration crude oil, coal, and natural gas', The Energy Journal, 27(2), pp. 55-71. Beck, R. and Solow, J. L. (1994) 'Forecasting nuclear power supply with Bayesian autoregression', Energy Economics, 16(3), pp. 185-192. BHP Billiton (2013) Annual Report. Available at: http://www.bhpbilliton.com/home/investors/reports/Documents/2013/BHPBillitonAnnualReport2013.pdf. BREE (2014) Resources and Energy Quarterly, Canberra: BREE. Available at: http://www.bree.gov.au/publications/resources-and-energy-quarterly. Brown, S. P. A. and Yücel, M. K. (2007) 'What drives natural gas prices?', Federal Reserve Bank of Dallas Working Paper. Bry, G. and Boschan, C. (1971) Cyclical Analysis of Time Series: Selected Procedures and Computer Programs. New York: National Bureau of Economic Research, Inc. Burns, A. F. and Mitchell, W. C. (1946) Measuring Business Cycles. New York: National Bureau of Economic Research. Cashin, P. and McDermott, C. J. (2002) 'The Long-Run Behavior of Commodity Prices: Small Trends and Big Variability', IMF Staff Papers, 49(2), pp. 175-199. Cashin, P., McDermott, C. J. and Scott, A. (2002) 'Booms and slumps in world commodity prices', Journal of Development Economics, 69, pp. 277-296. Chen, M., Clements, K. W. and Gao, G. 2012. Three Facts about World Metal Prices. Discussion Paper. The University of Western Australia Business School. Cuddington, J. T. and Urzúa, C. M. (1989) 'Trends and Cycles in the Net Barter Terms of Trade: A New Approach', The Economic Journal, 99(396), pp. 426-442. DFAT (2014) Nuclear Non-Proliferation, Trade and Security - Non Proliferation, Arms Control and Disarmament. Canberra: Australian Government Department of Foreign Affairs and Trade. Available at: www.dfat.gov.au/security/nuclear_safeguards.html (Accessed: 1 October 2014). DFATD (2014) Nuclear Cooperation Agreements. Ottawa: Government of Canada Department of Foreign Affairs, Trade and Development. Available at: www.international.gc.ca/arms-armes/nuclear-nucleaire/nca-acn.aspx?lang=eng (Accessed: 1 October 2014). Doane, M. J. and Spulber, D. F. (1994) 'Open access and the evolution of the US spot market for natural gas', Journal of Law and Economics, 37(2), pp. 477-517. EIA (2012) Annual Energy Review 2011, Washington, DC: US Energy Information Administration. EIA (2013) 2010 Manufacturing Energy Consumption Survey, Washington, DC: Energy Information Administration. EIA (2014) 2013 Uranium Marketing Annual Report, Washington, DC: US Energy Information Administration.

19

Engle, R. F. and Granger, C. W. J. (1987) 'Co-integration and error correction: representation, estimation, and testing', Econometrica, 55(2), pp. 251-276. ESA (2014) Annual Report 2013, Luxembourg: Euratom Supply Agency. European Commission (2014) Report on the Implementation of Euratom Safeguards in 2013, Brussels: European Union. Foss, M. M. (2005) 'Global natural gas issues and challenges: a commentary', The Energy Journal, 26(2), pp. 111-128. Gebre-Mariam, K. Y. (2011) 'Testing for unit roots, causality, cointegration, and efficiency: the case of the northwest US natural gas market', Energy, 36, pp. 3489-3500. Grilli, E. R. and Yang, M. C. (1988) 'Primary Commodity Prices, Manufactured Goods Prices, and the Terms of Trade of Developing Countries: What the Long Run Shows', World Bank Economic Review, 2(1), pp. 1-47. Hamilton, J. (2003) 'What is an oil shock?', Journal of Econometrics, 113, pp. 363-398. Hartley, P. R. 2013. The future of long-term LNG contracts. Discussion Paper. The University of Western Australia Business School. Hartley, P. R., Medlock, K. B. and Rosthal, J. E. (2008) 'The relationship of natural gas to oil prices', The Energy Journal, 29(3), pp. 47-65. Hotelling, H. (1931) 'The Economics of Exhaustible Resources', Journal of Political Economy, 39(2), pp. 137-175. IEA (2012) Policies for renewable heat, Paris: International Energy Agency. IEA (2013) Key World Energy Statistics, Paris: International Energy Agency. Ingram, S. R. (2015) 'Commodity Price Changes are Concentrated at the End of the Cycle', Credit and Capital Markets, 48(2), pp. 207-241. Johansen, S. (1988) 'Statistical analysis of cointegration vectors', Journal of Economic Dynamics and Control, 12(2), pp. 231-254. Kahouli, S. (2011) 'Re-examining uranium supply and demand: New insights', Energy Policy, 39, pp. 358-376. Labys, W. C., Kouassi, E. and Terraza, M. (2000) 'Short-term cycles in primary commodity prices', The Developing Economies, XXXVIII(3), pp. 330-342. Labys, W. C., Lesourd, J. B. and Badillo, D. (1998) 'The existence of metal price cycles', Resources Policy, 24(3), pp. 147155. Mjelde, J. W. and Bessler, D. A. (2009) 'Market integration among electricity markets and their major fuel source markets', Energy Economics, 31(3), pp. 482-491. Newbery, D. M. and Stiglitz, J. E. (1981) The theory of commodity price stablization. New York: Oxford University Press. OECD-NEA (2006) Forty Years of Uranium Resources, Production and Demand in Perspective, Paris: OECD Nuclear Energy Agency. OECD/IAEA (2006) Uranium 2005: Resources, Production and Demand, Paris: OECD Nuclear Energy Agency. OECD/IAEA (2010) Uranium 2009: Resources, Production and Demand, Paris: OECD Nuclear Energy Agency. OECD/IAEA (2012) Uranium 2011: Resources, Production and Demand, Paris: OECD Nuclear Energy Agency. OECD/IAEA (2014) Uranium 2014: Resources, Production and Demand, Paris: OECD Nuclear Energy Agency. Owen, A. D. (1985) The Economics of Uranium. Praeger Publishers: New York. Owen, A. D. (1992) 'Byproduct uranium', Resources Policy, June, pp. 137-147. Panagiotidis, T. and Rutledge, E. (2007) 'Oil and gas markets in the UK: evidence from a cointegrating approach', Energy Economics, 29, pp. 329-347. Prebisch, R. (1950) The Economic Development of Latin America and its principal problems. Lake Success, New York: United Nations Department of Economic Affairs. Radetzki, M. (1981) Uranium. London: Croom Helm Ltd. Roberts, M. C. (2009) 'Duration and characteristics of metal price cycles', Resources Policy, 34, pp. 87-102. Serletis, A. and Herbert, J. (1999) 'The message in North American energy prices', Energy Economics, 21, pp. 471-483. Serletis, A. and Rangel-Ruiz, R. (2004) 'Testing for common features in North American energy markets', Energy Economics, 26, pp. 401-414. Singer, H. W. (1950) 'The Distribution of Gains between Investing and Borrowing Countries', The American Economic Review, 40(2), pp. 473-485. Stigler, G. J. and Sherwin, R. A. (1985) 'The extent of the market', Journal of Law and Economics, 28(3), pp. 555-585. Taylor, J. H. and Yokell, M. D. (1979) Yellowcake: the international uranium cartel. USA: Pergamon Press. TradeTech (2014) TradeTech - Uranium Prices and Analysis since 1968. Colorado, USA. Available at: www.uranium.info (Accessed: 6 June. Trieu, L. H., Savage, E. and Dwyer, G. (1994) 'A model for the world uranium market', Energy Policy, 22(4), pp. 317-329. Villar, J. A. and Joutz, F. L. (2006) 'The relationship between crude oil and natural gas prices', Energy Information Administration. WNA (2014) World Nuclear Association. Available at: www.world-nuclear.org (Accessed: 12/03/2014.

20