The Long-term Influence of Fuel Prices on the
Movements of Primary Commodity Prices
Frances Houweling
1479849
June 2009
University of Groningen
Faculty of Economics and Business
The Long-term Influence of Fuel Prices on the
Movements of Primary Commodity Prices
FRANCES HOUWELING1
ABSTRACT
Do fuel prices influence the long-term movement of non-energy primary commodity prices? I study the worldwide impact of fuel prices on 40 primary commodities from 1970 to 2008. I use several cointegration techniques to estimate the influence of quarterly oil and natural gas prices on primary non-energy commodities. The Engle-Granger test results suggest that there is cointegration between the commodity and fuel prices in an isolated framework. The cointegration between the commodity and fuel prices ceases for some commodities, if several exogenous macroeconomic variables are added to the estimation with a Johansen VAR technique. I conclude that the influence of fuel prices on commodity prices in a macroeconomic environment depends on the individual commodity tested and cannot be generalized for all primary commodities. The implication is that several commodity prices are sensitive to fuel price movements and are likely to be influenced by fuel price shocks in the long run. On the methodological side, the results show that primary commodities should be studied on a separate basis, because aggregation in indices overlooks the individual commodity movement.
Keywords: primary commodities, cointegration, fuel prices, Johansen VAR, ECM JEL classification: E31, Q11, Q13, Q43
I. INTRODUCTION
rimary commodity prices have increased rapidly over the last seven years (graph 1). Since the summer of 2008 it would seem that the global economic crisis and the economic slowdown have, at least temporarily, put this surge to an end2. Although some experts say this decrease is the beginning of a worldwide slowdown, others believe it is a small distortion in the long commodity bull run3. Developing countries are most affected as they are reliant on commodity exports. The volatility of primary commodity prices causes major booms and slumps in these countries’ income and employment (Chaudhuri, 2001). Fuel prices, such as the price of crude oil and natural gas, have experienced extreme volatile price movements and have exerted an important influence on the world economy. For example, the increase in oil prices in 1974 and again in 1979 were important factors in producing a slowdown in the world economy, at a time when inflation was rising (Barrell and Pomerantz, 2004). A quantitative exercise carried out by the IEA (in collaboration with the OECD Economics Department and International Monetary Fund Research Department), showed that a sustained rise in oil prices, from $25 to $35, would result in the OECD, as a whole, losing 0.4 percent of GDP in the first two years following the increase. Inflation would rise by half a percentage point which would in turn increase unemployment4.
GRAPH 1
The long term movement of commodity and fuel indices
2New York Times, 24 October 2008 3Financial Times, 2 March 2009
4International Energy Agency (IEA), May 2004
The issue of explaining long-term movements in commodity prices is divisible into three streams of research: the structural approach, the Prebisch-Singer hypothesis and the excess co-movement hypothesis. The “traditional structural approach” relies exclusively on demand factors (Borensztein and Reinhart, 1994), to explain the primary commodity price behaviour. Demand side factors are, amongst others, industrial production, the exchange rate of the US Dollar and interest rates. These variables appeared to work well until 1984 when several industrial countries experienced a prolonged recession. Another line of research is the Prebisch-Singer hypothesis. This hypothesis is based on the historical fact that ever since the seventies, price trends have been heavily against sellers of food and raw materials and in favour of the sellers of manufactured goods. A common explanation for this phenomenon, is the observation that the income elasticity of demand for manufactured goods is greater than that for primary products - especially food. As incomes rise, therefore, the demand for manufactured goods increases more rapidly than the demand for primary products. Lastly, there is the excess co-movement hypothesis (Ai, Chatrath and Song, 2006; Deb, Trivedi and Varangis, 1996). This hypothesis is grounded on the co-movement in commodity price series (Pindyck and Rotemberg, 1990), meaning that different commodity prices exhibit, at least during some periods, a tendency to move together. Three main explanations for this phenomenon seem plausible. Supply and demand shocks in one commodity market may spill over into other markets. This spill-over effect is a logical explanation for commodities which are related to one another, either in production or consumption, but it cannot explain the co-movement between largely unrelated commodities (Tomek and Myers, 1993). Macroeconomic shocks may affect all prices together, but these shocks only explain a small fraction of the co-movement (Pindyck and Rotemberg, 1990). Speculators may overreact to new information and this causes spill-overs between commodity markets. In this interpretation excess co-movement amongst prices leads to volatility that is greater than it ought to be. The excess co-movement hypothesis (Ai, Chatrath and Song, 2006) states that the prices of commodities move together in a manner beyond what can be explained by fundamentals (macro-economic indicators such as inflation, industrial production and interest rates). This hypothesis calls into question the rationality of commodity markets and flies in the face of the competitive model of price formation. There is evidence (Pindyck and Rotemberg, 1990) which confirms this hypothesis and thus suggests there is irrational excess co-movement of prices.
commodities influence other primary commodities in several ways. The most logical reason for the inclusion of fuel prices, when forecasting commodity prices, is their role as input for the aggregate production function of the commodities (Chaudhuri, 2001; Borenzstein and Reinhart, 1994). This influence, as input, can be through energy intensive processes (transportation) or as a direct input (fertilizers). For some commodities there is the substitution effect of commodities other than oil and natural gas that can be used to produce ethanol (e.g. maize and sugar for ethanol production) or the substitution of commodities by products made from fuels such as synthetic rubber and man-made fibres (Baffes, 2007). Some countries are very dependent on the income they generate by producing fuel. Their prosperity can fluctuate with the fuel prices and influence the demand for certain commodities. On the other hand, the disposable income of non-fuel producing countries may be reduced when non-fuel prices rise, causing a decrease in industrial production and a reduced demand for raw materials and metals. The extraordinary events surrounding the first major oil price rise of 1973–74 may have generated a “limits to growth” psychological impact on other primary commodity markets (Gilbert and Perlman, 1987). Lastly, oil price spikes are often associated with inflationary pressures (Darby, 1892). Rising oil prices are expected to increase the prices of precious metals and other more tangible commodities as a safer way of storing wealth. Many energy experts believe natural gas will displace oil as the world's most widely used fuel. Consumption of natural gas, a cleaner and more efficient fuel, is rising in the US, Asia and Europe5. The major economies are now preparing for an increase in natural gas usage, due to concerns that oil supplies will fail to meet future energy needs6, to diversify their energy risk.
The impact of fuel prices on non-energy commodity prices is economically relevant, especially to developing countries. These countries depend on commodity exports, so price falls or fluctuations that affect them put exceptional strains on efforts aimed to reduce poverty (Page and Hewitt, 2001). If the non-energy primary commodity prices are influenced by the current volatile price movements or shocks in fuel commodities, the countries most affected are amongst the world’s poorest. Ninety-five of the 141 developing countries derive at least 50 percent of their export earnings from commodities (Brown and Gibson, 2006). For most of these countries their production represents only a small share of the competitive world markets and they therefore have no influence on price fluctuations (Page and Hewitt, 2001). The fluctuations in commodity prices can also be damaging to importers of commodities, especially food products. Low-income countries lack the resources to build up stock in periods of low prices or as insurance against price hikes (Page and Hewitt, 2001). Thus, when prices increase, they are asymmetrically affected. If fuel prices influence other commodities, developing countries
5EIA, International Energy Outlook 2008, June 2008
should diversify from this risk as much as possible. The research question I try to answer is: do fuel prices significantly influence non-energy primary commodity prices in the long run, independently of macroeconomic influences? I will consider this problem from the perspective of the governments of developing countries, from 1970 to 2008, in the world markets. The fuel commodities studied are petroleum and natural gas. The results of this study show that both petroleum prices and Russian natural gas prices are co-integrated with primary commodity prices. When I control for macroeconomic influences this long-term equilibrium disappears for some of the commodities. I conclude that the cointegration relationship between fuel and commodity prices in a macroeconomic framework cannot be generalized. Individual commodities show different interactions with fuel prices.
The structure of the remainder of this paper is as follows. A short outline of the literature on this topic is given in the next section. In section three, I will describe the data and its’ statistics and in section four I introduce the methodology. The results are presented and discussed in section five and finally I end with my conclusion.
II. REVIEW OF THE LITERATURE
commodities from 1960 to 2005. An OLS regression of the individual commodity price on the crude oil price is used, whilst taking inflation and technological changes into account. Baffes (2007) finds an elasticity for the non-energy commodity index of 0.16. When disaggregating this index the fertilizer index has the largest spill-over of 0.33. The second largest index is the agricultural commodity index with a coefficient of 0.17.
In this research I empirically examine the long-term impact of the different types of fuel prices on commodity prices for a prolonged period of time, including the most recent boom in commodity prices. To this extent, I will use the concept of cointegration. When the commodity prices and the fuel prices move together over time, they are co-integrated (Brooks, 2002, p388). That is, there is a long-term equilibrium relationship or, the two series trend together (Koop, 2008, p219). The co-integrated non-stationary variables can deviate in the short-term, but the cointegration stabilizes their relationship over time. The intuition behind cointegration is that although the prices of two products will fluctuate owing to the vagaries of supply and demand, market forces will always keep the price difference roughly constant (Koop, 2008, p219). This implies that when a commodity is co-integrated with a fuel price, the difference in price between the two will not in- or decrease very much. So when fuel prices surge, as is shown in graph 1, the commodity price is pushed up as well. I test if this cointegration relationship is still valid in the presence of other macroeconomic variables that might influence both individual variables as well as their mutual movement. Then, I test the following hypothesis: in the long run, commodity prices are co-integrated with fuel prices independently of the influence of macroeconomic factors. The null hypothesis being that no cointegration exists, if I take the macroeconomic variables into account.
This is the first study to address the long-term influence of fuel prices on individual primary commodities whilst taking into account macroeconomic variables. To the best of my knowledge gas prices have never been studied in such a framework before. Chaudhuri (2001) is the only researcher to study the cointegration between oil prices and individual primary commodity prices, however it represents an isolated case. Furthermore, I filter for macroeconomic impacts by treating them as exogenous variables using the Johansen VAR technique, whilst Hua (1998) includes them in an error correction model as endogenous variables. Finally, the time frame is very recent in this field of study.
III. DATA
Chaudhuri, 2001; Hua, 1998; Borensztein and Reinhart, 1994). Low frequency data are used to minimize the effect of crop cycles that agricultural commodities are subject to. Whilst a higher frequency would add more data points7, commodities are very volatile and the quarterly data may have larger information content8(Baffes, 2007). I use the IMF primary commodity index9 (2007) that consists of 44 non-energy primary commodities. Due to overlapping series I remove six commodities from this list. Furthermore, I add two fertilizers in order to compare the results with previous research (Baffes, 2007; Chaudhuri, 2001). This leaves 40 primary commodity price series, which are categorized as food, beverages, agricultural raw commodities, metals and fertilizers. The real primary commodity prices are expressed in US Dollars per metric ton or US Dollars per pound and deflated by world inflation (represented by the consumer prices in the US). A full description of all 40 non-energy commodities is given in table A.1. The petroleum price used is the worldwide crude petroleum price in US Dollars per barrel from the IFS (Baffes, 2007). This price is the average of the West Texas Intermediate, UK Brent and Dubai price. As the world market for natural gas is fragmented in different regional markets, it is not possible to refer to a world price for natural gas. Although there is a market liberalization trend all over the world, in many countries natural gas markets are still highly regulated10. As a result of different degrees of market regulation, natural gas prices differ among countries11. Two different variables for natural gas will therefore be tested, one for Europe and one for the US. The first natural gas price, a proxy for Europe, is estimated by the gas prices from the Russian Federation and is available from 1985. The second gas price is for US natural gas and this series starts in 1991. The correlation coefficient of the two natural gas series is 77.1 percent, for the petroleum price and Russian Natural Gas price 81.7 percent and lastly 79.9 percent for the petroleum price and the US Natural Gas price. The high correlation between petroleum and natural gas prices is expected as the fuels are related through both supply and demand (Villar and Joutz, 2006). Natural gas and petroleum prices are substitutes in consumption and complements, as well as rivals, in production. Villar and Joutz (2006) study the time series econometric relationship between the Henry Hub natural gas price and the West Texas Intermediate crude oil price and find a co-integrating relationship over the period 1989 through 2005. Their results show an asymmetric relationship: oil prices may influence the long-term development of natural gas prices, but are not influenced by them. Dvir and Rogoff (2009) note that in this field of research only studying the post-1973 period can be misleading. They argue that a long-term view on oil
7To test for robustness, I will also analyze monthly data. 8
In a comparison of US and Venezuelan income growth, Campos and Ericsson (1999) showed that 16 years of annual Venezuelan data carries almost twice the informational content than that of more than four decades of US quarterly data (162 observations).
9The IMF primary commodity index has 2002-2004 weights and is based on 2005=100. The index is prepared by Energy and Surveillance Unit.
10www.naturalgas.org/regulation
markets is appropriate, because shocks to the oil market have had different effects on the real price of oil during historical periods. The latter is due to the ability of key players in the market to restrict access to supplies.
TABLE 1
Descriptive statistic for all variables
Mean Median Max Min Std
Dev Obs
Average petroleum 25 19 121 2 20 156
Natural gas Russia 135 99 577 52 96 96
Natural gas US 146 97 444 47 94 72 Aluminum 1427 1381 3025 539 545 156 Bananas 456 431 913 224 143 136 Barley 87 79 239 39 32 136 Beef 98 101 136 52 21 156 Cocoa Beans 1706 1583 4066 487 739 156
Coffee, other milds 114 116 288 44 48 156
Coffee, robusta 87 78 269 22 47 156 Copper 2293 1761 8454 1019 1517 156 Cotton 66 66 111 28 17 156 DAP 202 178 1192 53 147 156 Fish 5 5 8 3 2 119 Fishmeal 688 653 1270 315 228 156 Groundnuts 792 770 2415 215 335 156 Hides 62 68 100 11 24 156 Iron ore 33 28 141 12 23 156 Lead 683 546 3220 229 473 156 Lamb 110 112 185 32 36 156 Maize 110 108 259 49 32 156 Nickel 8201 6071 47764 2795 6965 156 Olive oil 3210 2980 6167 1814 1202 121 Oranges 500 450 1322 170 215 136 Palm oil 402 378 1089 172 157 156 Poultry 55 55 88 31 14 115 Rice 289 273 953 123 114 156 Rubber 46 40 139 14 23 156 Shrimp 11 12 19 3 4 156 Soybean meal 202 196 401 89 57 156 Soybeans 241 229 490 165 55 79 Soybean oil 495 478 1346 189 171 156 Sugar 20 21 48 7 6 156 Sunflower oil 607 584 2231 258 255 156 Superphosphate TSP 162 138 1108 40 129 155 Swine meat 79 70 170 27 29 115 Tea 196 197 389 99 53 156
Timber hard logs 161 161 493 38 82 156
Timber hard sawnwood 429 469 942 92 241 156
Tin 8212 6662 22544 3486 4039 156
Uranium 20 14 122 7 20 115
Wheat 148 148 411 52 50 156
Wool 432 423 840 142 138 156
Exchange rate 1 1 2 1 0 156
Avg ind production 76 72 118 44 20 156
Interest rate 7 6 19 1 4 156
NOTE: For units see table A.1
Table 1 gives the descriptive statistics for all variables used. I find that the absolute standard deviation gives high values for nickel, tin, copper and olive oil. The macroeconomic variables show a low absolute standard deviation. When looking at the standard deviation relative to the mean values, uranium, nickel and superphosphate TSP show values above 0.8. US natural gas is the shortest time series with 72 observations.
Macroeconomic indicators are generally seen as the principal factor affecting commodity prices (Hua, 1998). Boughton and Branson (1988) and Frankel (1986) show that expectations concerning macroeconomic disturbances have an important role in the price formation process of commodities. I use the industrial production (IP), the real effective Dollar exchange rate (ER) and the interest rate (IR) as macroeconomic variables. Industrial production is closely related to the demand for primary commodities. This is due to the fact that a rise in industrial production will directly increase the demand for raw materials and intermediate inputs and indirectly raise the demand for food and beverages through a rise in incomes (Reinhart and Wickham, 1994; Borensztein and Reinhart, 1994; Hua, 1998; Ai, Chatrath and Song, 2006). Real interest rates represent the opportunity costs of holding commodities as a portfolio asset (Hua, 1998). The costs of carry are higher when interest rates increase. High interest rates can also pressure commodity prices through the substitution effect of reducing prices (Hua, 1998), via the rising of debt service obligations (Gilbert, 1989) and the flight of capital from developing countries (Rausser et al., 1990). Lastly, the exchange rate of the US Dollar against a portfolio of the currencies of other countries may directly influence the commodity prices by affecting demand in these countries (Hua, 1998). Real commodity prices and the real exchange rate are jointly determined (Borensztein and Reinhart, 1994). Pindyck and Rotemberg (1990) also use the S&P 500 Stock index as a variable, but do not find a significant impact on the commodity prices studied. Additionally, oil prices from October 1973 until 1979 were largely dictated by the OPEC12cartel. The OPEC cartel caused a worldwide crisis from 1973 to 1974 when they proclaimed an oil embargo in response to the Israeli conflict. Since OPEC lost control of the market, prices have been primarily determined by market forces. A dummy variable therefore, is unity from 1973q1 to 1974q2 to correct for the extraordinary behaviour at the time of this second OPEC price embargo.
The order of integration (d) is the minimum number of differences required to obtain a stationary series. If a variable is perfectly stationary it has a constant mean, constant variance and constant autovariances for each given lag. If a series is I(1), it contains a unit root which can lead to a spurious regression. To test for a unit root in the level of the commodities I use the Augmented Dickey Fuller test with a trend and intercept included in the equation (Chaudhuri, 2001). The lag length is automatically selected with the Schwartz Info Criterion, with a maximum of 13. Table A.3 contains the results from the Augmented Dickey Fuller tests (ADF) (Dickey and Fuller, 1979). The results suggest that for 25 of the variables the null hypothesis of non-stationarity is rejected at a 5 percent significance level. However, the ADF test statistic is well known for its low power if the test is stationary, but with a unit root close to the non-stationary boundary (Brooks, p381). To take this lack of power into account I consider a confirmatory data analysis and test for a null hypothesis of stationarity. This test, developed by Kwiatkowski et al. (1992), shows non-stationarity for different commodities other than the ADF test statistics. Lastly, I also use the Philips-Perron test (Phillips and Perron, 1988). This test is similar to the ADF test, but in addition makes an automatic correction to allow for autocorrelated residuals. The test suffers from the same limitations as the ADF test statistic. The results in table A.3 show that only palm oil, poultry and soybean meal are stationary on a 5 percent significance level for all three test statistics. These three stationary commodities can be categorised under the IMF food index. I will assume for the remainder of this research that all other variables are integrated to the order one.
IV. METHODOLOGY
In order to test the long-term influence of fuel prices on 40 primary commodity prices in a macroeconomic environment, I first need to ensure that a long run relationship between the fuel prices and the individual commodity prices exists. To estimate this relationship I use the Engle-Granger (1987) test of cointegration. Secondly, the long-term influence of fuel prices on commodity prices is estimated together with the short-term properties of the relationship using an error correction model (ECM). The last step is to establish that this relationship holds in a macroeconomic framework. This hypothesis is examined by re-estimating the ECM, in addition to applying the Johansen VAR technique. This approach is in line with Hua (1998), Chaudhuri (2001) and Palaskas and Varangis (1989).
Engle-Granger (1987) test of cointegration between the individual primary commodity prices and the different fuel prices. If cointegration is present, the spurious regression problem vanishes and the unit roots in Ptand FUELt‘cancel each other out’ (Koop, 2008, p218). The Engle-Granger cointegration test is a two-step procedure, also used by Chaudhuri (2001). Firstly, a static OLS regression is estimated, represented by the equation:
t t
t FUEL
P
0
1
(1)Secondly, I test the residuals from equation 1 on their stationarity using the Augmented Dickey Fuller test (Chaudhuri, 2001). It is worth mentioning that the Engle-Granger procedure is based on a unit root test, so the problems with the low power of the test, as described in the previous section, will arise. For the Engle-Granger test to be valid the variables should be non-stationary at an absolute level, but stationary at the same differenced level (Engle-Granger, 1987). As shown in table A.3, the hypothesis of non-stationarity cannot be rejected for all the variables under consideration except palm oil, poultry and soybean meal.
If a long-term relationship has been established between a pair or set of variables, a dynamic error correction model of the relationship always exists (Palaskas and Varangis, 1989). Palaskas and Varangis (1989) show with an ECM that macroeconomic variables have good forecasting abilities on commodity prices. The error correction model allows the long-term components of the commodity prices to obey equilibrium constraints, whilst their short-term components have a flexible dynamic adjustment (Hua, 1998). I estimate two ECM’s, one including the macroeconomic variables and one excluding these influences. The models are estimated with an intercept in the same way as Hua (1998) and Palaskas and Varangis (1989). I cannot estimate both models for all non-fuel commodities, this would preclude a meaningful interpretation of the results on the individual commodity level, due to the production of an unmanageable amount of data. I therefore make a selection of commodities whereby each subcategory is represented by one commodity. The commodities are selected on the basis of the following criteria:
The availability of the time series data for the whole sample period (table 1) The presence of a unit root in the series (table A.3)
The cointegration relationship with all three different fuels (to enable a cross-fuel analysis)
After applying the first and second criteria, 29 commodities remain. From these 29 commodities a selection is made, based on the cointegration relationship I find in the results. These criteria ensure that the ECM is only estimated between variables that are all I(0) and co-integrated. The two variable, one lag, reduced form error correction specification can be written as:
where Δ denotes rate of change, Ptthe price of the individual commodity and ECMt is the
residual of the cointegration regression (lnPt
0
1lnFUELt
t). The multivariable, one lag reduced form error correction model is represented by:t t t t t t t t t D ECM IR ER IP FUEL P P
1 6 5 4 3 2 1 1 0 ln ln ln ln ln ln (3) in this equation ECMt is the residual of the cointegration regression with macroeconomicvariables (lnPt
0
1lnFUELt
2lnIPt
3lnERt
4lnIRt
t).The Engle-Granger method suffers from three problems (Brooks, 2002, p394). Firstly, as discussed earlier, there may be a lack of power in the unit root and cointegration tests. Secondly, there may be a simultaneous equations bias if the causality runs in both directions. This two-step procedure with a single equation forces one to make one variable dependent and one independent, even if there is no theoretical rationale for doing so. Finally, it is not possible to perform any hypothesis tests regarding the actual cointegration relationship estimated between the commodity price and the fuel price tested. Johansen (1988), developed another technique which overcomes the latter two problems. The Johansen technique is based on maximum likelihood estimation of a vector autoregressive system (VAR) and used in the studies of Hua (1998) and Chaudhuri (2001). This co-integrating system allows me to take the effect of macroeconomic influences as exogenous factors into account. The first step is to specify a VAR model of the order 4, with Gaussian errors:
t t t t t t c y y y y y
1 1
2 2
3 3
4 4
(4) where yt = (lnPt, lnFUELt) is a 2 x 1 vector of variables that are I(1), c is a 2 x 1 vector ofconstants and εtis a 2 x 1 vector of errors. This VAR can then be rewritten into a vector error correction model (VECM) of the form
t t t t t i i t i t t c y y IP ER IR D y
1 2 3 4 3 1 4 ln ln ln (5) where g i i I
) ( 4 1
; ( ) 1
i j ji
and I is the identity matrixThe vectors Γicapture the short term disequilibrium features of the data, the vector Φ contains
the macroeconomic influences for the different variables and εt is a vector of errors. The
non-zero Eigenvalues of the matrix (r) (Brooks, 2002, p404). Under the hypothesis of cointegration, Π has a reduced rank r < 2 and can be written as
'
(6)
β is a 2 x r matrix of cointegration vectors and α is a 2 x r matrix giving the amount of each co-integrating vector entering each equation of the VECM, known as the ‘adjustment parameters’ (Brooks, 2002, p406).
Under the Johansen approach (1988) there are two different likelihood ratio test statistics: the trace test and maximum Eigenvalue test, shown in equations (7) and (8).
) ˆ 1 ln( 2 1
r i i trace T
(7) and ) ˆ 1 ln( 1 max T
r
(8)where T is the sample size and
ˆiis the estimated value of the ith ordered Eigenvalue from the Π matrix. The trace test tests the H0 of r co-integrating vectors against the H1 of two co-integrating vectors. The maximum Eigenvalue test tests the same H0 against the alternative hypothesis of r+1 vectors. The critical values used when estimating these statistics with Eviews are from Osterwald-Lenum (1992).V. RESULTS
In this section I will discuss the results from the Engle-Granger cointegration tests, the ECM, the Johansen VAR technique and the robustness tests, as specified in the previous section. For each of these tests I will first examine the econometric results in relation to previous literature and then give an economic interpretation. The results for the Engle-Granger cointegration test are presented in table 2. The commodities are sub categorized by the latest version of the IMF primary commodity index (2007). The energy commodities are not included in this analysis. For each commodity three co-integrating relations are estimated; petroleum average, natural gas Russia and natural gas US. The ADF statistic and the corresponding lags are given in table A.3. If the null hypothesis of non-stationary of the regression’s residuals is rejected, it can be concluded that there is cointegration between the variables.
TABLE 2
Engle-Granger cointegration test for 40 non-energy commodities
t t
t FUEL
P 01 , testing with an ADF test for a unit root in the error term εt Lags Petroleum average Lags Natural Gas Russia Lags Natural Gas US EDIBLES Beverages Cocoa Beans 1 -2.852 1 -3.141 0 -1.943
Coffee, other milds 1 -3.666** 0 -2.708 0 -2.142
Coffee, Robusta 3 -3.591** 1 -2.525 1 -2.158 Tea 0 -4.205** 0 -5.433** 1 -2.442 Food – Cereals Maize 1 -4.575** 2 -3.227* 1 -3.092 Wheat 0 -4.051** 1 -4.654** 0 -2.280 Rice 1 -3.568** 1 -4.448** 0 -1.870 Barley 0 -3.378* 1 -3.281* 1 -2.829 Food - Vegetable Oil and Protein Meals
Sugar 1 -5.235** 1 -4.782** 1 -4.798** INDUSTRIAL INPUT
Agricultural raw materials
Timber hard logs 2 -2.543 2 -2.544 9 0.233
Timber hard sawnwood 0 -2.005 1 -2.288 3 -2.002 Cotton 1 -4.033** 1 -3.502** 1 -3.285* Wool 3 -5.110** 1 -2.704 0 -2.203 Rubber 0 -3.031 1 -2.534 0 -2.365 Hides 0 -1.733 0 -4.088** 1 -4.131** Metals Aluminum 1 -3.457** 1 -2.610 0 -2.658 Copper 0 -2.913 7 -4.173** 1 -3.112 Iron ore 3 -1.412 4 -4.459** 0 -1.082 Lead 1 -5.481** 3 -6.024** 10 -1.323 Nickel 5 -3.598** 3 -4.175** 1 -3.454* Tin 1 -2.662 3 -5.257** 3 -2.911 Uranium 4 -2.034 1 -4.464** 6 -0.765 Zinc 2 -5.516** 2 -3.697** 1 -2.795 Fertilizers DAP 1 -4.489** 4 -5.297** 1 -3.564** Superphosphate TSP 1 -1.910 3 -4.367** 0 3.532
NOTE: * = 10% and ** = 5% significance level. The statistic is the Augmented Dickey Fuller statistic (Dickey and Fuller, 1981) which tests the null hypothesis of non-stationarity of the level series.
whilst the US market is highly liberalized allowing prices to respond to supply and demand forces. After their liberalization the US gas prices declined significantly. In the Russian Federation there is a clear monopoly and domestic gas prices are kept artificially low whilst internationally the gas is sold for a higher price to recover losses. The diverging results between the two natural gas series may also be due to the shorter time series available for the US prices. For the US natural gas variable there are 72 data points available, which makes it difficult to extract stable results13.
As described in the methodology a commodity is selected for each sub-category of the IMF primary commodity index. The following commodities: tea, wheat, soybean oil, lamb, sugar, cotton, nickel and diammonium phosphates (DAP) meet the requirements described in the methodology. For these commodities the two variable ECM will be estimated. Table 3 shows the results for this ECM as specified in equation 2. The table gives the regression coefficients and t-values for each OLS estimation. The two-variable model captures a considerable part of the commodity behaviour. The estimated equations explain between 15 and 25 percent of the adjusted primary commodity variance. The R2is significantly lower only for lamb, with 5.3 percent for the average petroleum equation. These values for R2 are low if compared to the results from Baffes (2007). When Baffes (2007) regresses the oil price and a price deflator on the individual commodity prices he finds an adjusted R2 around 0.7. The estimates for the ECM show a positive significant coefficient for the short-term growth rate of the commodity prices in their precedent level (ΔlnPt-1). A positive autocorrelation in the commodity returns
can thus be demonstrated, implying that in the short-term, prices are dependent on their previous growth rate. The commodity price will be adjusted with approximately 0.2 to 0.5 percent of the previous year’s growth rate. In contrast, tea has a significant negative coefficient for the precedent growth rate. This means that for tea a short-term price correction mechanism exists. The long-term denominator ECMt-1 is very significant and negative for all
combinations. This result indicates that if the difference between the lags of the commodity and the fuel prices is positive in one period, the commodity price may fall during the next period to restore equilibrium. The adjustment to this equilibrium is below 20 percent of the previous year’s disequilibrium for most commodities, in line with the coefficients Hua (1998) finds. This finding supports the hypothesis of long-term cointegration of commodity prices and the different fuel prices. The last point to discuss is the influence of the short-term fuel growth rate (ΔlnFUELt). Six of the eight commodities are positively influenced by a change in the
current petroleum price, on a 10 percent significance level. If the oil price rises, commodity prices are likely to follow and increase around ten percent of the price surge.
17 TABLE 3
Two variable error correction model for eight non-energy commodities
t t t t t P FUEL ECM P
ln 0 1 ln 1 2 ln 3 1 , where ECMtis the residual of the regression; lnPt 01lnFUELt t
COEFFICIENTS (t-value)
Average petroleum Natural Gas Russia Natural Gas US
β0 ΔlnPt-1 ΔlnOILt ECMt-1 β0 ΔlnPt-1 ΔlnNGEUt ECMt-1 β0 ΔlnPt-1 ΔlnNGUSt ECMt-1
Tea 0.002 0.121 0.108 -0.264 -0.001 0.019 0.212 -0.364 0.005 -0.209 0.035 -0.239 (0.235) (1.525) (1.806)* (-4.919)** (0.102) (0.188) (2.040)** (-4.069)** (0.411) (-1.692)* (0.601) (-2.526)** Wheat 0.004 0.309 0.105 -0.159 0.003 0.383 -0.019 -0.153 0.004 0.342 0.017 -0.136 (0.461) (3.700)** (1.972)** (-3.826)** (0.272) (3.611)** (-0.204) (-3.151)** (0.273) (2.674)** (0.271) (-2.577)** Soybean oil 0.001 0.332 0.149 -0.185 -0.002 0.480 -0.026 -0.163 -0.001 0.506 0.025 -0.132 (0.080) (4.039)** (2.502)** (-4.426)** (-0.229) (4.466)** (-0.276) (-3.542)** (-0.056) (3.755)** (0.429) (-2.867)** Lamb 0.007 0.170 0.043 -0.065 0.005 0.266 -0.058 -0.086 0.002 0.371 0.050 -0.243 (1.176) (2.042)** (1.141) (-2.160)** (0.791) (2.538)** (-0.921) (-2.025)** (0.288) (3.286)** (1.461) (-3.763)** Sugar 0.002 0.469 0.076 -0.154 0.000 0.266 -0.040 -0.382 0.000 0.345 -0.009 -0.445 (0.291) (6.730)** (1.635) (-4.815)** (0.059) (2.482)** (-1.190) (-4.798)** (-0.072) (2.738)** (-0.391) (-4.337)** Cotton 0.001 0.347 0.092 -0.134 -0.003 0.345 0.087 -0.194 -0.007 0.297 0.068 -0.172 (0.143) (4.343)** (1.763)* (-3.845)** (-0.293) (3.354)** (0.896) (-3.708)** (-0.582) (2.387)** (1.178) (-2.858)** Nickel 0.001 0.372 0.151 -0.074 0.001 0.458 0.054 -0.078 -0.004 0.506 0.039 -0.105 (0.057) (4.523)** (2.197)** (-2.704)** (0.070) (4.325)** (0.342) (-1.930)* (-0.201) (4.082)** (0.442) (-2.219)** DAP 0.002 0.376 0.270 -0.116 0.006 0.430 0.005 -0.129 0.002 0.476 0.139 -0.090 (0.241) (4.650)** (4.638)** (-3.428)** (0.509) (3.554)** (0.048) (-2.778)** (0.120) (3.304)** (2.060)** (-2.096)**
The natural gas series do not have a similar short-time influence on the primary commodity price level. Russian natural gas only influences tea in the short term and US natural gas shows a significant coefficient for DAP. The constant in all estimations is very close to zero and insignificant. This lack of significance is expected since the residuals are estimates of errors, which have a mean zero.
Overall, these results indicate that the prices of primary commodities and the prices of the different energy sources move together in the long run. This implies that the recent commodity price decrease is likely to continue if oil and gas prices keep falling. To minimize exposure to these influences developing countries should try and filter energy costs out of the prices. The adjustment of commodity prices to a long-term equilibrium with fuel prices is slow. In my view, this slow adjustment might be caused by long-term (negotiated) price contracts for both the fuels and the non-energy primary commodities. The contracts freeze the price over a specified period of time and the adjustment of the prices will then take more than one lag. The multivariate ECM, the Johansen VAR and the robustness analysis, will be estimated using the results for petroleum and Russian natural gas only. The time series for the US natural gas is too short to establish a meaningful long-term relationship.
TABLE 4
Multivariable error correction model with the average petroleum price
t t t t t t t t t P OIL IP ER IR ECM D P ln 0 1 ln 1 2 ln 3 ln 4 ln 5 ln 6 1
where ECMtis the residual of the regression: t t t t t t OIL IP ER IR P ln ln ln ln ln 0 1 2 3 4 Coefficients (t-value)
β0 ΔlnPt-1 ΔlnOILt ΔlnIPt ΔlnERt ΔlnIRt ECMt-1 Dt
Tea -0.001 0.021 0.098 0.373 0.446 -0.015 0.007 0.018 (-0.088) (0.258) (1.453) (2.490)** (1.272) (-0.197) (0.176) (0.316) Wheat 0.004 0.185 0.043 0.242 -0.213 0.047 -0.059 0.043 (0.497) (2.124)** (0.766) (1.925)* (-0.741) (0.746) (-1.775)* (0.864) Soybean oil -0.003 0.153 0.070 0.037 -0.107 0.019 -0.121 0.179 (-0.340) (1.832)* (1.106) (0.261) (-0.328) (0.268) (-3.251)** (3.214)** Lamb 0.004 0.179 0.001 0.123 0.623 -0.051 -0.020 0.027 (0.656) (2.208)** (0.026) (1.443) (3.196)** (-1.186) (-0.893) (0.846) Sugar -0.001 0.397 0.041 -0.129 0.038 0.072 -0.041 0.117 (-0.113) (5.362)** (0.806) (-1.126) (0.143) (1.236) (-1.332) (2.657)** Cotton 0.000 0.250 0.059 -0.054 0.032 0.038 -0.057 0.060 (-0.056) (2.947)** (1.043) (-0.435) (0.113) (0.600) (-1.711)* (1.248) Nickel 0.003 0.319 0.137 -0.027 -0.194 0.021 -0.078 -0.015 (0.260) (3.756)** (1.855)* (-0.167) (-0.524) (0.258) (-1.842)* (-0.243) DAP -0.002 0.438 0.228 -0.495 0.233 -0.085 -0.178 0.129 (-0.229) (5.537)** (3.952)** (-3.740)** (0.787) (-1.295) (-4.968)** (2.641)**
The strong cointegration relationship in table 3 changes, if the equation with the macroeconomic variables, as described in equation 3, is re-estimated. Table 4 shows that the long-term equilibrium relation is no longer significant for tea, sugar and lamb. The other commodities show significantly negative coefficients for the cointegration factor, in line with the significantly negative residual coefficients Hua (1998) finds for the price indices. For only one commodity, tea, the ECMt-1sign is surprisingly positive; if the difference between
commodity and fuel prices is positive in one period, the commodity price will rise even more in the next period and the gap will widen. This positive coefficient is in contrast with Hua’s (1998) results for the beverage price index. Hua (1998) and Palaskas and Varangis (1989) also find that the interest rate influences the commodity prices, whereas the growth in the interest rate is not significant in table 4. The remaining macroeconomic variables have scattered positive significant influence. One strange outlier in the results for the macroeconomic variables is the negative influence of the industrial production on DAP. In other words, if there is a worldwide growth in industrial production, the price of this fertilizer will fall in the short-term. Finally, in the short run, prices are very dependent on their preceding growth rates.
20 TABLE 5
The trace and maximum Eigenvalue statistics for the Johansen VAR cointegration technique
Average Petroleum Natural Gas Russia
Trace Max Eigenvalue Trace Max Eigenvalue
H0no. of CE Statistic H0no. of CE Statistic H0no. of CE Statistic H0no. of CE Statistic
Tea None ** 28.837 None ** 22.505 None * 15.819 None 11.209 At most 1 * 6.332 At most 1 * 6.332 At most 1 * 4.611 At most 1 * 4.611
Wheat None ** 33.706 None ** 27.323 None ** 24.807 None ** 20.848 At most 1 * 6.382 At most 1 * 6.382 At most 1 * 3.958 At most 1 * 3.958
Soybean oil None ** 29.963 None ** 22.747 None ** 24.642 None ** 21.450 At most 1 ** 7.216 At most 1 ** 7.216 At most 1 3.192 At most 1 3.192
Lamb None ** 30.466 None ** 23.162 None ** 33.101 None ** 28.553 At most 1 ** 7.304 At most 1 ** 7.304 At most 1 * 4.547 At most 1 * 4.547
Sugar None 13.521 None 9.558 None * 18.414 None * 16.438 At most 1 * 3.964 At most 1 * 3.964 At most 1 1.975 At most 1 1.975
Cotton None ** 29.368 None ** 22.173 None ** 28.574 None ** 25.940 At most 1 ** 7.195 At most 1 ** 7.195 At most 1 2.633 At most 1 2.633
Nickel None ** 31.872 None ** 19.745 None * 17.630 None 13.469 At most 1 ** 12.127 At most 1 ** 12.127 At most 1 * 4.162 At most 1 * 4.162
DAP None ** 33.931 None ** 26.961 None * 18.870 None * 15.822 At most 1 ** 6.969 At most 1 ** 6.969 At most 1 3.048 At most 1 3.048
disaggregated individual commodity data, taking into account that the different commodities give very different results. This suggested research goes beyond the scope of this study, where the focus is on the influence of fuel prices on commodities.
Table 5 shows the results for the Johansen VAR technique. Whilst analyzing these results I emphasize the maximum Eigenvalue test, because the trace test tends to have more heavily distorted sizes (significance levels) (Gregory, 1994; Lütkepohl, Saikkonen and Trenkler, 2001). Thus, where the two tests show different results, the maximum Eigenvalue test takes precedence. The results for petroleum in table 5 show rejection of both hypotheses (no cointegration and one unit root), on a 5 percent level, in all cases except for sugar. There are two implications for these results. Firstly, this result could indicate that the variables are stationary, but as shown in table A.3 this stationarity is unlikely. Secondly, the VAR system could be overfitted and this may cause the rejection of both hypotheses. When analyzing the table on a 1 percent significance level, tea and wheat are co-integrated. For sugar the hypothesis of a unit root is rejected on a 5 percent level. This is an unexpected result, since sugar is the main input of a direct substitute (ethanol) for petroleum. A possible cause for the lack of significance is that the ethanol production has gained renewed interest in the United States since 2000 (Coyle, 2007). It is plausible that it did not exert significant influence over the period 1970 to 2008. I test this assumption in the robustness analysis. For petroleum no conclusions can be drawn based on these findings. The results do imply that even with exogenous influences, in the long run, non-energy commodity prices are sensitive to fuel prices. Russian natural gas is significantly co-integrated with 6 out of the 8 commodities. Tea and nickel have no long-term relationship with the natural gas price according to these results, contradicting the multivariable ECM model for natural gas. Due to the commodity-specific nature of the results, future studies should analyse the data as disaggregately as possible, preferably on the individual commodity level. Furthermore, the results are sensitive to different testing methodologies and statistical form. The most contradictory results are obtained from the multi-variable ECM model where the macroeconomic series are defined as endogenous variables.
the commodities selected for the ECM are not robust. This result confirms the aforementioned need to study commodity data on a disaggregate level. If an overall index is studied (Hua, 1998; Baffes, 2008) the individual commodity movement is overlooked whilst these can differ significantly as the estimations show. The Engle-Granger results are robust for the use of monthly data.
TABLE 6
Summary of the main results for the sub sample sensitivity analysis
Engle-Grangera Johansen VARb
Average
Petroleum Natural GasRussia PetroleumAverage Natural GasRussia
First 78 data pointsc 8 3 1 na
Second 78 data pointd 11 15 1 na
Stable period e 19 7 0 0
Increasing period f 5 9 2 7
NOTE: the number in the table is the number of commodities that are integrated with this sub sample only. a Total number of commodities
tested is 40. b Total number of commodities tested is 8. cFrom Q1 1970 to Q2 1989. dFrom Q3 1989 to Q4 2008. eFrom Q1 1970 to Q4
2001. fFrom Q1 2002 to Q4 2008. Source: table A.5 to A.8.
cointegration factor is negative and significant for all commodities in line with table 3. The short-term autocorrelation coefficient is significant on a 5 percent level for all commodities, except beef. The short-term oil price coefficient is not significant for the beverage- and cereal-categories, whilst they are significant in table 3. Finally, the multivariable model only shows cointegration for coffee, with a significant positive coefficient for the interest rate in line with Palaskas and Varangis (1989). These results stress the need to work with individual commodities, as each individual commodity shows different results for the cointegration tests in a macroeconomic environment.
VI. CONCLUSION
I started my thesis by questioning whether fuel prices significantly influence non-energy primary commodities in the long run, independently of macroeconomic influences. I find that the influence of fuel prices on primary commodity prices depends on the individual commodity tested. This result is in line with the research done by Baffes (2007). To answer the research question I investigate the hypothesis of cointegration of commodity prices with fuel prices, independently of the influence of macroeconomic factors in the period of 1970 to 2008. I use the Engle-Granger cointegration model, the error correction model and the Johansen VAR technique for quarterly data. The results of the Engle-Granger model show that in the long run the non-stationarity of primary commodity prices can be attributed to the non-stationarity of petroleum and natural gas prices. Food and beverages are significantly cointegrated with petroleum, as are metals with Russian natural gas. In the short-term the commodity prices are autocorrelated with their previous growth rates and petroleum prices have a positive influence on the different commodity spot prices. These findings are in line with the results Chaudhuri (2001) finds and suggest that commodity prices will track fuel prices if a shock happens. The Johansen technique shows that this cointegration relationship disappears for some individual commodities if several macroeconomic indicators are added as exogenous variables to the framework. This implies that the cointegration relationship encompasses the entire macroenvironment instead of solely the fuel price. For commodities such as tea cointegrated with petroleum or cotton with Russian natural gas, the null hypothesis of no cointegration of commodity prices with fuel prices, independently of the influence of macroeconomic factors, can be rejected. A sensitivity analysis shows that the recent price surge significantly affects the relationship between fuel prices and commodity prices.
commodity prices as independent of fuel prices as possible. There are two ways in which this can be achieved: diversify the commodities exported or diversify the energy structure of the country. To diversify within the commodities exported it is important for the individual countries to know the movements of the individual commodities as well as the influence of the individual commodities on the export portfolio of the specific country. Governments of these countries should try to find their optimal portfolio of commodities, given certain external constraints. These constraints are factors such as world prices and the ability to produce certain commodities on a large scale. The second solution is to diversify the energy structure of the individual country. This solution is derived from the result that different fuels influence commodities in various ways. Furthermore, alternative sources of energy, such as solar, wind or water, can be developed in the country itself so as to become self-sufficient in its energy supply. These suggestions limit the influence of fuel shocks on commodity prices but are unfortunately not short-term solutions.
This is the first study to address the influence of fuel prices on individual commodity prices whilst taking the macroeconomic environment into account. My results emphasize the need to study commodities on an individual basis. Besides, it is the first time that the influence of natural gas prices on primary commodities has been verified. This research has several limitations which should be taken into consideration. Due to limited sources the data set is rather short and not complete for all variables. For example, the Russian natural gas series only starts in 1985 instead of the preferred 1970. Dvir and Rogoff (2009) note that studying petroleum prices only for the post-1973 period can be misleading. To add to this, the error correction model and Johansen VAR technique are not estimated for all the individual commodities and the results are sensitive to statistical form. Further research could include additional or different variables to filter for macroeconomic influences. Furthermore, it may be interesting to design a more complete VAR system, to test the relationships amongst all variables and to use more lags when testing with an error correction model.
REFERENCES
Ai, C., A. Chatrath and F. Song (2006). On the Comovement of Commodity Prices, American Journal of Agricultural Commodities, 88, 574-588
Baffes, J. (2007). Oil Spills on Other Commodities, World Bank Working Paper, August
Barrell, R. and O. Pomerantz (2004). Oil Prices and the World Economy, National Institute Economic and Social Research, Discussion Paper 242
Boughton, J.M. and W.H. Branson (1988). Commodity Prices as a Leading Indicator of Inflation, NBER Working Paper Series, No. 2750
Brooks, C. (2002). Introductory Econometrics for Finance, Cambridge University Press, Cambridge, UK
Brown, O. And J. Gibson (2006). Boom or Bust: Developing countries’ rough ride on the commodity price rollercoaster, International Institute for Sustainable Development, Winnipeg, Canada Burbidge, J. and A. Harrison (1984). Testing for the Effects of Oil-price Rises using Vector
Autoregressions, International Economic Review, 25, 459-484
Campos, J. and N.R. Ericsson (1999). Constructive data mining: modeling consumers’ expenditures in Venezuela, Econometrical Journal, 2, 226-240
Chaudhri, K. (2001). Long-run Prices on Primary Commodities and Oil Prices, Applied Economics, 33, 531-538
Coyle, W. (2007). The Future of Biofuels: A Global Perspective, Amber Waves, 5, 24-29
Darby, M.R. (1982). The Price of Oil and World Inflation and Recession, The American Economic Review, 72, 738-751
Deb, P., P.K. Trivedi and P. Varangis (1996). The Excess Co-Movement of Commodity Prices Reconsidered, Journal of Applied Econometrics, 11, 275-291
Dickey, D. and W.A. Fuller (1979). Distribution of the Estimators for Time Series Regressions with Unit Roots, Journal of the American Statistical Association, 74, 427-431
Dvir, E. and K.S. Rogoff (2009). Three Epochs of Oil, NBER Working Paper Series, No. 14927 Engle, R.F. and C.W.J. Granger (1987). Cointegration and error correction: representation, estimation
and testing, Econometrica, 55, 251-276
Frankel, J.A. (1986). Expectations and Commodity Price Dynamics: The Overshooting Model, American Journal of Agricultural Economics, 68, 344-348
Gilbert, C.G. (1989). The impact of exchange rates and developing country debt on commodity prices, Economic Journal, 99, 773-84
Gilbert C.G. and R.A. Perlman (1987). The Prospect for Commodities, The World Day, 43, 4-7 Gregory, A.W. (1994). Testing for Cointegration in Linear Quadratic Models, Journal of Business &
Economic Statistics, 12, 347-360
Hua, P. (1998). On Primary Commodity Prices: The Impact of Macroeconomic/Monetary Shocks, Journal of Policy Modeling, 20, 767-790
International Monetary Fund (2007). IMF Primary Commodity Indices: Comparison of Old and New Indices, IMF Commodities Unit, Washington DC
Johansen, S. (1988). Statistical Analysis of Cointegration Vectors, Journal of Economic Dynamics and Control, 21, 637-677
Kwiatkowski et al. (1992). Testing the Null Hypothesis of Stationarity against the Alternative of a Unit Root, Journal of Econometrics, 54, 159-178
Lütkepohl, H., P. Saikkonen and C. Trenkler (2001). Maximum Eigenvalue versus Trace Test for the Cointegration Rank of a VAR Process, The Econometrics Journal, 4, 287-310
Osterwald-Lenum, M. (1992). A Note with Quantiles of the Asymptotic Distribution of the Maximum Likelihood Cointegration Rank Test Statistics, Oxford Bulletin of Economics and Statistics, 54, 461-472
Page, S. and J. Hewitt (2001). World commodity prices: Still a problem for developing countries?, Overseas Development Institute, London, UK
Palaskas, T. and P. Varangis (1989). Primary Commodity Prices and Macroeconomic Variables, Working Papers, WSP 314, Washington DC: World Bank
Phillips, P.C.B. and P. Perron (1988). Testing for a Unit Root in Time Series Regression, Biometrica, 65, 335-346
Pindyck, R.S. and J. J. Rotemberg (1990). The Excess Co-Movement of Commodity Prices, Economic Journal, 100, 1173-1189
Reinhart, C. and P. Wickham (1994). Time Series Properties of Commodity Prices, International Monetary Fund, 41, 506-524
Rausser, G.C., M.B. Rose and D.A. Irwin (1990). World Commodity Prices: The Role of External Debt and Industrial Country Policies, Agricultural Trade Liberalisation, Implication for Developing Countries (I. Goldin and O. Knudsen, Eds.), Paris: OECD; Washington DC: World Bank, pp 488
Tomek W.G. and R.J. Myers (1993). Empirical Analysis of Agricultural Commodity Prices: a viewpoint, Review of Agricultural Economics, 15, 181-202
APPENDIX TABLE A.1 Primary commodities
Series Units Region Code
Average Petroleum U.S. dollars per barrel World 00176AAZZF... Natural Gas Russia Russia 92276NGZZF... Natural Gas US United States 11176NGZZF... Aluminum U.S. dollars per metric ton Canada 15676DRZZF... Bananas U.S. dollars per metric ton Ecuador 24876U.ZZF... Barley U.S. dollars per metric ton Canada 15676BAZZF... Beef U.S. cents per pound Australia 19376KBZZF... Cocoa Beans U.S. dollars per metric ton Ghana 65276R.ZZFM44 Coffee, other milds U.S. cents per pound Central America 38676EBZZF... Coffee, robusta U.S. cents per pound Africa 79976ECZZF... Copper U.S. dollars per metric ton United Kingdom 11276C.ZZF... Cotton U.S. cents per pound United States 11176F.ZZFM40 DAP U.S. dollars per metric ton United States 11176ARZZKM17
Fish Norway 14276FIZZF...
Groundnuts U.S. dollars per metric ton Nigeria 69476BHZZF... Hides U.S. cents per pound United States 11176P.ZZF...
Iron ore Brazil 22376GAZZF...
Lead U.S. dollars per metric ton United Kingdom 11276V.ZZF... Lamb U.S. cents per pound New Zealand 19676PFZZF... Maize U.S. dollars per metric ton United States 11176J.ZZFM17 Nickel U.S. dollars per metric ton Canada 15676PTZZF... Olive oil U.S. dollars per metric ton United Kingdom 11276LIZZF... Oranges U.S. dollars per metric ton France 13276RAZZF... Palm oil U.S. dollars per metric ton Malaysia 54876DGZZF... Poultry U.S. cents per pound United States 11176POZZF... Rice U.S. dollars per metric ton Thailand 57876N.ZZFM81 Rubber U.S. cents per pound Malaysia 54876L.ZZF... Shrimp U.S. dollars per pound United States 11176BLZZF... Soybean meal U.S. dollars per metric ton United States 11176JJZZF... Soybeans U.S. dollars per metric ton Brazil 22374S.ZZF... Soybean oil U.S. dollars per metric ton United States 11176JIZZF... Sugar U.S. cents per pound United States 11176IAZZFM02 Sunflower oil U.S. dollars per metric ton United Kingdom 11276SOZZF... Superphosphate TSP U.S. dollars per metric ton United States 11176ASZZF... Swine meat U.S. cents per pound United States 11176SMZZF...
Tea United Kingdom 11276S.ZZF...
Timber hard logs Malaysia 54876VXZZF... Timber hard sawnwood Malaysia 54876RMZZF... Tin U.S. dollars per metric ton United Kingdom 11276Q.ZZF... Uranium U.S. dollars per pound World 00176UMZZF... Wheat U.S. dollars per metric ton United States 11176D.ZZF...
Zinc U.S. dollars per metric ton United Kingdom 11276T.ZZF...
NOTE: the codes are IFS time series codes. These codes provide information about the time series that supplements the text descriptors. More information about the content of the codes can be found on http://www.imfstatistics.org.proxy-ub.rug.nl/imf/IFSCodes.htm. Source: IMF, International Financial Statistics.
TABLE A.2 Macroeconomic variables Variable Description and source
IPt The average industrial production index for 24 advanced economies1equally weighted and
non-seasonally adjusted. Constructed by the author. Source: IMF, International Financial Statistics, 11066..IZF...
ERt IMF end of period market exchange rate of the United States Dollar per SDR2. Calculated by
applying these countries GDP weights to their corresponding nominal dollar exchange rates, corrected by the consumer prices in United States and other industrial countries (Hua, 1998). Source: IMF, International Financial Statistics, 111..AA.ZF...
IRt Nominal interest rate represented by the 90 days London interbank offer rates on the United
States dollar deposits, reduced by the change of the consumer prices in the United States (Hua, 1998). The period averages are in percent per annum. Source: IMF, International Financial Statistics, 11160LDDZF...
Dt The variable is unity from 1973q1 to 1974q2 to correct for the extraordinary behaviour at the
time of the second OPEC price embargo.
NOTE: 1) The advanced economies consist of: Korea, New Zealand, Australia, Spain, Portugal, Malta, Ireland, Greece, Finland, Japan, Canada, Switzerland, Sweden, Norway, Netherlands, Luxembourg, Italy, Germany, France, Denmark, Belgium, Austria, United Kingdom and the United States. 2) The currency value of the SDR is determined by summing the values in U.S. dollars, based on market exchange rates, of a basket of major currencies (the U.S. dollar, Euro, Japanese yen, and pound sterling). The SDR currency value is calculated daily and the valuation basket is reviewed and adjusted every five years (IMF).
TABLE A.3
Unit root tests for 43 primary commodities
ADFa Obs Lags KPSSb PPc
Oil -2.031 156 2 0.204** -2.529
Natural Gas Russia 1.549 96 3 0.283*** 2.308
Natural Gas US -4.885*** 72 1 0.211** -3.242* Aluminium -3.544** 156 1 0.104 -3.141 Bananas -1.853 136 3 0.120* -7.676*** Barley -4.020*** 136 1 0.176** -3.077 Beef -2.518 156 0 0.188** -2.629 Cocoa Beans -2.809 156 1 0.143* -2.223
Coffee, other milds -3.612** 156 1 0.140* -3.173*
Lamb -3.623** 156 1 0.096 -3.258* Maize -4.581*** 156 1 0.094 -3.279* Nickel -3.387* 156 1 0.153** -3.269* Olive oil -3.649** 121 3 0.056 -2.628 Oranges -1.395 136 2 0.203** -5.071*** Palm oil -4.620*** 156 1 0.102 -3.458** Poultry -5.021*** 115 4 0.069 -3.972** Rice -3.883** 156 1 0.097 -3.012 Rubber -4.195*** 156 9 0.127* -2.725 Shrimp -2.197 156 0 0.322*** -1.749 Soybean meal -5.373*** 156 1 0.083 -4.077*** Soybeans 0.656 79 0 0.156** -0.263 Soybean oil -4.746*** 156 1 0.104 -3.154* Sugar -5.307*** 156 1 0.189** -3.354* Sunflower oil -4.707*** 156 1 0.148** -3.491** Superphosphate 1.010 155 5 0.155** 3.155 Swine meat -4.343*** 115 0 0.166** -4.546*** Tea -4.070*** 156 0 0.148** -4.121***
Timber hard logs -2.754 156 2 0.119 -3.242*
Timber hard sawn
wood -2.720 156 1 0.111 -2.398 Tin -1.914 156 2 0.162** -1.670 Uranium -2.560 115 1 0.227*** -1.997 Wheat -4.025*** 156 1 0.103 -3.179* Wool -4.681*** 156 3 0.131* -3.111 Zinc -4.548*** 156 2 0.085 -3.294*
NOTE: * = 10%, ** = 5%, *** = 1% significance level. aADF is the Augmented Dickey Fuller statistic
(Dickey and Fuller, 1981) which tests the null hypothesis of nonstationarity of the level series. The critical values are from MacKinnon (1996).bKPSS tests the null hypothesis of stationarity (Kwiatkowski et al., 1992). c PP is the Phillips-Perron test statistic (Phillips and Perron, 1988) which tests the null hypothesis of
TABLE A.4
Multivariable error correction model with the Russian natural gas price
t t t t t t t t t P NGEU IP ER IR ECM D P ln 0 1 ln 1 2 ln 3 ln 4 ln 5 ln 6 1
where ECMtis the residual of the regression; t t t t t t NGEU IP ER IR P ln ln ln ln ln 0 1 2 3 4
NOTE: * = 10% and ** = 5% significance level.
Coefficients (t-value)
β0 ΔlnPt-1 ΔlnNGEUt ΔlnIPt ΔlnERt ΔlnIRt ECMt-1
TABLE A.5
Sensitivity analysis Engle-Granger cointegration test for 40 non-energy commodities
t t
t FUEL
P 01 , testing with an ADF test for a unit root in the error term εt
Petroleum average Natural Gas Russia
First 78 Lags Second 78 Lags First 78 Lags Second 78 Lags EDIBLES
Beverages
Cocoa Beans 0.570 1 0.287 1 0.785 0 0.065 1
Coffee, other milds 0.132 1 0.439 0 0.228 1 0.428 0
Coffee, robusta 0.485 4 0.769 0 0.234 1 0.392 1 Tea 0.029 3 0.095 2 0.004 1 0.011 0 Food - Cereals Maize 0.232 0 0.593 2 0.862 0 0.119 2 Wheat 0.397 0 0.147 0 0.922 0 0.003 1 Rice 0.069 1 0.431 0 0.660 1 0.016 0 Barley 0.172 0 0.383 0 0.306 1 0.034 1 Food - Vegetable Oil and Protein Meals
Soybean meal 0.011 1 0.052 1 0.603 0 0.003 1 Soybeans na na 0.954 0 na na 0.808 1 Soybean oil 0.193 0 0.156 1 0.036 1 0.000 1 Palm oil 0.184 0 0.106 1 0.012 1 0.005 1 Sunflower oil 0.159 1 0.092 1 0.724 0 0.001 1 Olive oil 0.463 0 0.058 3 0.040 0 0.299 3 Groundnuts 0.000 3 0.002 1 0.116 1 0.000 1 Food - Meat Beef 0.226 0 0.633 0 0.148 0 0.603 0 Lamb 0.168 0 0.082 1 0.123 1 0.264 1 Swine meat 0.215 0 0.001 4 0.614 0 0.001 4 Poultry 0.472 0 0.005 1 0.619 0 0.837 5 Food - Seafood Shrimp 0.404 0 0.152 0 0.753 0 0.164 0 Fish 0.078 7 0.820 2 0.997 0 0.795 2 Food - Other Bananas 0.882 4 0.000 0 0.926 2 0.000 0 Oranges 0.542 6 0.000 0 0.005 2 0.491 2 Sugar 0.005 1 0.001 1 0.051 0 0.001 1 INDUSTRIAL INPUT Agricultural raw materials
Timber hard logs 0.356 0 0.197 1 0.267 0 0.119 1
Timber hard sawnwood 0.631 0 0.542 1 0.181 0 0.108 3
Copper 0.548 0 0.348 0 0.562 0 0.022 7 Iron ore 0.075 0 0.995 2 0.043 3 0.001 4 Lead 0.755 0 0.001 1 0.035 0 0.000 3 Nickel 0.041 3 0.000 3 0.673 0 0.251 6 Tin 0.358 0 0.354 1 0.487 1 0.000 4 Uranium 0.037 1 0.744 4 1.000 3 0.009 1 Zinc 0.188 1 0.004 2 0.759 0 0.062 2 Fertilizers DAP 0.061 1 0.298 5 0.719 0 0.000 4 Superphosphate TSP 0.037 1 0.981 1 0.288 2 0.000 3
NOTE: The table gives the probability of the Augmented Dickey Fuller statistic (Dickey and Fuller, 1981) which tests the null hypothesis of non-stationarity of the level series.
TABLE A.6
Sensitivity analysis Engle-Granger cointegration test for 40 non-energy commodities
t t
t FUEL
P 01 , testing with an ADF test for a unit root in the error term εt Petroleum average Natural Gas Russia
Stable Lags Surge Lags Stable Lags Surge Lags EDIBLES
Beverages
Cocoa Beans 0.218 1 0.930 0 0.668 0 0.258 4
Coffee, other milds 0.043 1 0.733 2 0.458 0 0.987 0
Coffee, robusta 0.030 3 0.184 0 0.474 1 0.080 5 Tea 0.006 0 0.100 2 0.001 0 0.035 2 Food - Cereals Maize 0.010 1 0.790 0 0.123 1 0.066 5 Wheat 0.099 0 0.326 0 0.545 0 0.397 4 Rice 0.019 1 0.778 0 0.341 0 0.782 4 Barley 0.112 0 0.800 0 0.531 0 0.038 5 Food - Vegetable Oil and Protein Meals
Bananas 0.023 9 0.006 1 0.000 0 0.005 1
Oranges 0.081 2 0.036 0 0.133 2 0.177 4
Sugar 0.000 1 0.013 1 0.033 1 0.012 1 INDUSTRIAL INPUT
Agricultural raw materials
Timber hard logs 0.333 0 0.046 1 0.558 0 0.138 3
Timber hard sawnwood 0.618 1 0.000 3 0.910 0 0.264 3
Cotton 0.022 1 0.545 0 0.214 1 0.240 1 Wool 0.082 1 0.487 1 0.023 4 0.310 2 Rubber 0.649 0 0.276 0 0.646 1 0.803 1 Hides 0.402 0 0.536 0 0.051 0 0.479 0 Metals Aluminum 0.136 1 0.000 5 0.144 1 1.000 2 Copper 0.317 0 0.058 3 0.254 1 1.000 0 Iron ore 0.047 0 0.120 3 0.693 0 0.044 4 Lead 0.184 1 0.024 1 0.430 0 0.007 3 Nickel 0.013 3 0.018 3 0.005 3 0.044 5 Tin 0.170 0 0.582 1 0.000 1 0.022 5 Uranium 0.058 3 0.664 4 0.470 1 0.530 1 Zinc 0.072 1 0.002 3 0.270 0 0.799 2 Fertilizers DAP 0.007 6 0.990 0 0.556 0 0.008 5 Superphosphate TSP 0.003 1 1.000 0 0.241 1 0.059 4
