Assessing the Validity of the Oil-Agri
Commodity Price Nexus
Master’s thesis
MSc Economic Development & Globalisation
University of Groningen
Faculty of Economics and Business
Author: Owen Nicolaas de Boer Student Number: S2924706 Mail address: o.n.de.boer@student.rug.nl
Supervisor: Prof. Dr. J. de Haan Co-assessor: Dr. M.V. Nikolova
ABSTRACT
This study analyses the long-run relationship between the price dynamics of world crude oil markets and global agricultural commodity markets by employing the concepts of cointegration and Granger causality through the ARDL bounds and Toda-Yamamoto causality tests. The empirical analysis utilises the daily futures prices of the Brent and WTI crude oil contracts as well as those of ten key agricultural commodity contracts, covering the period of the 2nd of
January 1990 till the 18th of September 2020. The findings affirm the presence of a relationship
with the price dynamics of world crude oil markets for several of the agricultural commodities in the period of 2004-2020. Contrasting with the existing literature, the results indicate the price dynamics of these commodity markets to be interlinked through a channel other than the cost-push effect, biofuels or aggregate demand. The financialisation of commodity markets is proposed as an alternative explanation.
Keywords: oil prices, agricultural commodity prices, commodities futures, cointegration, causality
List of Figures
FIGURE 1:PRICE GRAPH OF BRENT &WTI CRUDE OIL ... 42
FIGURE 2:PRICE GRAPH OF BRENT CRUDE OIL &GOLD ... 43
FIGURE 3:PRICE GRAPH OF BRENT CRUDE OIL &CORN ... 44
FIGURE 4:PRICE GRAPH OF BRENT CRUDE OIL &WHEAT ... 45
FIGURE 5:PRICE GRAPH OF BRENT CRUDE OIL AND SOYBEAN ... 46
FIGURE 6:PRICE GRAPH OF BRENT CRUDE OIL AND SOYBEAN OIL ... 47
FIGURE 7:PRICE GRAPH OF BRENT CRUDE OIL AND SUGAR ... 48
FIGURE 8:PRICE GRAPH OF BRENT CRUDE OIL AND RICE ... 49
FIGURE 9:PRICE GRAPH OF BRENT CRUDE OIL AND ARABICA COFFEE ... 50
FIGURE 10:PRICE GRAPH OF BRENT CRUDE OIL AND ROBUSTA ... 51
FIGURE 11:PRICE GRAPH OF BRENT CRUDE OIL AND COCOA ... 52
FIGURE 12:THE LINKAGES BETWEEN THE PRICES OF CRUDE OIL AND AGRICULTURAL COMMODITIES ... 53
FIGURE 13:GLOBAL BIOFUEL PRODUCTION FROM 2000-2021. ... 53
List of Tables
TABLE 1:RESULTS OF THE GRAPHICAL ANALYSIS OF THE COMMODITY PRICE SERIES:1990-2004 ... 17TABLE 2:RESULTS OF THE GRAPHICAL ANALYSIS OF THE COMMODITY PRICE SERIES:2004-2020 ... 18
TABLE 3:DESCRIPTIVE STATISTICS OF THE COMMODITY PRICE SERIES:1990-2004 ... 18
TABLE 4:DESCRIPTIVE STATISTICS OF THE COMMODITY PRICE SERIES:2004-2020 ... 19
TABLE 5:CORRELATION MATRIX ... 20
TABLE 6:UNIT ROOT TEST WITH STRUCTURAL BREAK:1990-2004 ... 28
TABLE 7:UNIT ROOT TEST WITH STRUCTURAL BREAK:2004-2020 ... 28
TABLE 8:ARDL BOUNDS TEST WITH STRUCTURAL BREAK:1990-2004 ... 29
TABLE 9:ARDL BOUNDS TEST WITH STRUCTURAL BREAK:1990-2004 ... 30
TABLE 10:ARDLERROR CORRECTION REGRESSION:1990-2004 ... 31
TABLE 11:ARDL BOUNDS TEST WITH STRUCTURAL BREAK:2004-2020 ... 31
TABLE 12:ARDL BOUNDS TEST WITH STRUCTURAL BREAK:2004-2020 ... 33
TABLE 13:ARDLERROR CORRECTION REGRESSION:2004-2020 ... 34
TABLE 14:TODA-YAMAMOTO CAUSALITY TEST:1990-2004 ... 35
TABLE 15:TODA-YAMAMOTO CAUSALITY TEST:2004-2020 ... 36
TABLE 16:UNIT ROOT TESTS WITHOUT STRUCTURAL BREAK: LEVELS ... 54
TABLE 17:UNIT ROOT TESTS WITHOUT STRUCTURAL BREAK: FIRST DIFFERENCES ... 55
CONTENT
1. INTRODUCTION ... 5
2. LITERATURE REVIEW ... 8
2.1. THEORY ... 8
2.2. EMPIRICAL LITERATURE ... 9
2.3. HYPOTHESES &RELEVANCE ... 12
2.3.1. Hypotheses ... 12 2.3.2. Relevance ... 13 3. DATA ... 15 4. PRELIMINARY ANALYSIS ... 17 4.1. PRICE GRAPHS ... 17 4.2. DESCRIPTIVE STATISTICS ... 18 4.3. CORRELATION MATRIX ... 20 5. METHODOLOGY ... 21
5.1. UNIT ROOT TESTS ... 21
5.2. ARDL BOUNDS TESTS ... 22
5.3. TODA-YAMAMOTO CAUSALITY TESTS ... 25
6. EMPIRICAL FINDINGS & DISCUSSION ... 27
6.1. UNIT ROOT TESTS ... 27
6.1.1. Standard unit root tests ... 27
6.1.2. Unit root test with structural break ... 27
6.2. ARDL BOUNDS TESTS &ECM ... 29
6.2.1. ARDL bounds test without structural break ... 29
6.2.2. ARDL bounds test with structural breaks: 1990-2004 ... 29
6.2.3. Error correction model: 1990-2004 ... 30
6.2.4. ARDL bounds test with structural breaks: 2004-2020 ... 31
6.2.5. Error correction model: 2004-2020 ... 33
6.3. TODA-YAMAMOTO CAUSALITY TESTS ... 34
6.4. DISCUSSION ... 35
7. CONCLUSION, LIMITATIONS & FUTURE RESEARCH ... 39
7.1. CONCLUSION ... 39
7.2. LIMITATIONS ... 40
7.3. FUTURE RESEARCH ... 41
ANNEX I – PRICE GRAPH ANALYSIS ... 42
ANNEX II – THE OIL-AGRICULTURAL COMMODITY PRICE NEXUS ... 53
ANNEX III – GLOBAL BIOFUEL PRODUCTION 2000-2021 ... 53
ANNEX IV – STANDARD UNIT ROOT TEST RESULTS ... 54
ANNEX V – RESULTS ARDL BOUNDS TEST WITHOUT STRUCTURAL BREAKS. ... 56
ANNEX VI – CUSUM & CUSUMSQ TEST RESULTS ... 57
ANNEX VII – DATASET ... 76
1. Introduction
Throughout time, commodity markets have occupied a pivotal role in sustaining economic growth as well as improving human welfare (Morse & Clayton, 2015). As such, the functioning of commodity markets has always been a point of interest to scholars, policy makers and market participants. However, as a result of the pandemic, commodity markets have found themselves in the limelight once more as lockdowns, supply disruptions and the slowing of the global economy rattled the world’s commodity markets (World Bank, 2020).
Although several noteworthy events can be singled out, one of the most prominent has been the crash of the oil price. This April, oil prices hit a historic low as demand dropped in response to the global slowdown in production and mobility at a time where prolonged uncertainty around the production levels persevered (Engebretsen, 2020; Smith, Razzoul & Martin, 2020). Although no longer trading below zero, the pandemic is expected to have lasting impact on the world price of crude oil due to structurally altering consumer and employment behaviour (Baffes & Nagle, 2020). A second event worth singling out has been the subjugation of agricultural commodity markets to not only a temporary rise in protectionism, but also to a true shopping spree by countries and citizens alike. Accordingly, it is within this environment that agricultural prices, along with food insecurity and malnutrition, have risen globally (Baffes & Nagle, 2020; Almeida & Murtaugh, 2020).
Through these events, the pandemic has re-emphasized the importance of commodity markets for the global economy and human welfare. Herewith, it has underlined the added value of understanding the key drivers behind the price dynamics of commodity markets. Of particular significance is, in light of the current as well as predicted future low oil price environment, the relationship between the price dynamics of crude oil and agricultural commodity markets. The prices of agricultural commodities, although dependent on a large multitude of intricate and interdependent factors, are believed to be driven for a significant part by the price of crude oil (Nazlioglu, 2011).
Although the empirical evidence is conclusive as to the presence of a relationship between the price dynamics of world crude oil markets and global agricultural commodity markets, it is far from consensual in respect to the existence and importance of the different linkages. To exemplify, whereas Baffes (2007) and Gohin and Chantret (2010) find evidence supportive of the cost-push effect, Ciaian and Kancs (2011) conclude the cost-push effect to be small and statistically insignificant. Nevertheless, a similar case can be made for the biofuel channel. There, the existing literature, despite being consensual as to the relevance of the biofuel channel in linking the price of crude oil with those of soybean and corn, finds mixed evidence for sugar and wheat (Zafeiriou, Arabatzis, Karanikola, Tampakis & Tsiantikoudis, 2018; Saghaian, 2010; Ciaian & Kancs, 2011). Likewise, the role of aggregate demand has also been subject to debate. To illustrate, Baumeister and Killian (2014) as well as Hochman, Rajagopal and Zilberman (2010) find aggregate demand to be an explanation for the relationship between the price of crude oil and several agricultural commodities from 2003-2008. However, the World Bank (2014) concludes there to be little evidence of a systematic shift in the consumption of agricultural commodities by emerging economies. Furthermore, Baffes and Etienne (2004) report the income elasticity of agricultural commodities to be small. To put it all in a nutshell, the existing literature has been unable to present a conclusive answer as to the presence or the importance of the different channels.
Accordingly, to further our knowledge on the linkages between the price dynamics of the world crude oil and global agricultural commodity markets, this study analyses the presence of a long-run relationship between the world crude oil price and global agricultural commodity prices as well as its origin(s). This has been formalised in the following research question: “To what extent is there a long-run cointegrated relationship between the world price of crude
oil and global agricultural commodity prices and if so, can its presence be attributed to the cost-push effect, the biofuel channel and/or aggregate demand?”
To analyse the relationship and its origin(s), this study applies the concepts of cointegration and Granger causality by utilising the Auto-Regressive Distributed Lag (ARDL) bounds and Toda-Yamamoto causality test. The data comprises of daily futures prices of Brent crude oil, WTI crude oil, gold, corn, wheat, soybean, soybean oil, sugar, rice, Robusta coffee, Arabica coffee and cocoa for the period of the 2nd of January 1990 to the 18th of September 2020. To
account for the disrupting impact of the rapid expansion of biofuel production on the price dynamics of commodity markets, the studied time period is split into two separate time periods using the 26th of December 2003 as a cut-off date. This distinction does not only isolate the
cost-push effect in the pre-biofuel era, but it also helps scrutinising the impact of the introduction of biofuels on the relationship between the price dynamics of oil and agricultural markets.
Previous research has often confined itself to either analysing the linkages between commodity markets in general, or by limiting their study on the linkages between the price dynamics of agricultural and oil markets to a meagre selection of agricultural commodities. Neither of these approaches lend themselves well to adequately research the linkages between the price dynamics of world crude oil markets and those of global agricultural commodity markets. In contrast, this study offers a more comprehensive analysis by including a broad selection of agricultural commodities. Last but not least, from 2015 onwards little research has been done within this field. Thus, by including data up to the 18th of September 2020, this study captures
not only the longest time period to date, but it also accounts for a time period which has been researched relatively little.
The findings of this study indicate there to be no long-run relationships, or Granger causality, between the world price of crude oil and the global agricultural commodity prices for the period 1990-2004. Alternatively, for 2004-2020 long-run relationships are observed between the price of crude oil and the prices of soybean, soybean oil, rice, Arabica coffee and cocoa. The majority of these results are consolidated by those of the Toda-Yamamoto causality test. As the explanations put forward by the literature are unable to explain several of the observed long-run relationships, the findings point towards the presence of an alternative linkage between the price dynamics of the world crude oil markets and the global agricultural commodity markets. The financialisation of commodity markets is put forward as a plausible and well-fitting explanation.
2. Literature Review
Over time, the interdependencies between world oil and global agricultural markets have received increasing attention as the volatility and price hikes from oil markets have appeared to coincide more and more with those of the already volatile agricultural markets (Saghaian, 2010; Nazlioglu, 2011). To understand how the dynamics of crude oil markets may be linked with those of agricultural commodity markets, it is important to discuss the various channels through which these commodity markets may be related. Accordingly, this chapter will begin by summarising the main theoretical explanations for the interdependencies of these markets. Thereafter, the empirical findings of previous scholars are discussed to illustrate the current state of the academic debate. The last section of this chapter is dedicated to the hypotheses and outlines this study’s contribution to the literature.
2.1. Theory
Within the existing literature there three main explanations for the similarities between the price dynamics of oil and agricultural markets. Namely, the cost-push effect, biofuels, and aggregate demand (Vo, Vu, Vo & McAleer, 2019).
The traditional channel through which the price dynamics of oil markets appear to impact those of agricultural markets, is through the cost-push effect. The cost-push effect rationalises that changes in the price of crude oil, through various energy intensive inputs such as fertilizer and fuel, influence the prices of agricultural commodities (Baffes, 2007). This effect is believed to be strengthened by the importance of long-distance transportation, which is common for many agricultural commodities given not only their global production but also their global consumption (Baumeister & Kilian, 2014). Vice versa, this channel also allows for the possibility of agricultural markets impacting oil markets as efficiency gains within the agricultural sector, lessening the dependence on energy intensive inputs, lower the demand for crude oil (Ciaian & Kancs, 2011). The size of the cost-push effect hinges on the relative importance of energy intensive inputs to the total costs of production (Koirala et al., 2015). For example, the cultivation of wheat, soybeans, corn, rice and sugar comes with a high dependence on energy intensive inputs due to mechanised farming as well as large scale use of fertilizers. To illustrate, the share of total operating expenses of US farmers from 2007-2008 attributable to energy-related inputs exceeded 50% for wheat, corn and rice. For soybeans, energy related costs were comparatively low at about 35% (Sands & Westcott, 2011). Alternatively, cocoa and coffee require little energy intensive inputs. For example, cocoa is mainly produced by poor smallholders, who rely on low input cultivation systems. As such, cocoa is farmed using forest soil fertility and manual labour (Wessel & Quist-Wessel, 2015). Similarly, the production of coffee is highly dependent on manual labour, although mechanised farming is slowly increasing. To illustrate, within Colombia, Costa Rica and Honduras, 75%, 57% and 56% of total production costs derive from manual labour. As for fertilizers, although their use is widespread among farmers, their share of total production costs is comparatively low (ICO, 2016; ICO, 2019).
dependence and to shift towards more renewable energy sources (Ajanovic, 2011). Biofuels, encompassing bioethanol and biodiesel, are fuels produced from biomass. In turn, biomass predominantly comprises of agricultural commodities such as corn, soybeans, sugarcane and wheat. To exemplify, about 64% of the world’s ethanol is produced from corn, 26% is based on sugarcane and an additional 3% on wheat. Similarly, 77% of the world’s biodiesel originates from vegetable oils such as rapeseed oil (37%) and soybean oil (27%) (OECD, 2020). Consequently, given the importance of corn, soybeans, sugarcane and wheat to the production of biofuels, changes in the price of crude oil are likely to not only affect agricultural commodity prices through the cost-push effect, but also through changes in the demand for biofuels (Ajanovic, 2011; Nazliogu, 2011). To illustrate, higher crude oil prices will result in higher gasoline prices and accordingly, higher biofuel prices. In turn, these will lead to a scale-up of the production of biofuels, increasing not only the demand for commodities such as corn, soybeans, sugarcane and wheat, but also their prices. However, the biofuel channel is also likely to result in spill-over effects as farmers tend to allocate their arable land towards biofuel related commodities when prices are high, reducing the available supply of other commodities and therewith increasing their prices (Saghaian, 2010).
Last but not least, the similarities between crude oil and agricultural commodity price dynamics have also been explained through the rapid economic growth of China, India and other emerging economies (Tang & Xiong, 2012). Their income growth, is argued to have contributed to an increase in aggregate demand for crude oil as well as agricultural commodities, causing both prices to rise simultaneously, ceteris paribus (Shahzad, Hernandez, Al-Yahyaee & Jammazi, 2018; Baumeister & Kilian, 2014; Hochman, Rajagopal & Zilberman, 2010).
In summary, three explanations have been suggested for the similar price dynamics of world crude oil and global agricultural markets. Namely, the cost-push effect, biofuels, and income growth of emerging economies. For illustrative purposes, these have been visualised in figure 12 in annex II.
2.2. Empirical literature
As a result of the intricate linkages between the world crude oil and global agricultural markets, the existing literature does not provide conclusive evidence for the existence and/or size of the different linkages. Previous research on the relationship between the price of crude oil and agricultural commodity prices has relied upon a large variety of statistical methods, ranging from multiple linear regression to ARMA GARCH models. Nonetheless, the majority has relied on VAR, ARDL and VEC models along with the complementary cointegration and causality tests (Wang, Wu & Yang, 2014).
assess the spill-over effects of changes in the price of crude oil on corn, wheat, soybeans and rice, provide further support for the presence of a long-run relationship between all agricultural commodity prices considered and the price of crude oil.
A limitation of the studies discussed above is the absence of any further analysis in respect to the origin of the reported long-run relationships. Contrasting herein is the study by Baffes (2007), who focusses on quantifying the cost-push effect. Through the application of multiple linear regression, using annual price indices of several key commodities, Baffes finds a cost-push effect for cocoa, corn, wheat, soybean and soybean oil. A similar result is reported by Gohin and Chantret (2010), who also attribute the relationship found between world crude oil and food prices to the cost-push effect. Likewise, the study conducted by Koirala et al. (2015), utilising Copula models and daily futures price data for commodities such as crude oil, corn and soybeans, finds a cost-push effect. In contrast, Ciaian and Kancs (2011), who analyse the transmission of changes in the oil price on the prices of corn, wheat, rice, sugar, soybeans, cotton, bananas, sorghum and tea over the period of 1994-2008, conclude that agricultural commodity prices are only cointegrated with the price of crude oil over the period 2004-2008. As such, no indications of cointegration between these prices was found for the period of 1994-2003. These results, further consolidated by the results of Granger Causality tests, are argued to point towards an insignificantly small cost-push effect. Consequently, Ciaian and Kancs attribute the cointegration found between 2004-2008 to the presence of a biofuel channel with a strong spill-over effect.
threshold cointegration to analyse whether co-movement between corn, soybeans and soybean oil prices occurs once oil prices exceed a certain threshold. The results suggest that such is indeed the case for corn, as corn-prices are found to co-move with the price of crude oil after the latter has exceeded the price of $75 per barrel. Natanelov et al. (2011) explain their findings through the impact of policy interventions. Namely, subsidies in the US have kept the price of corn artificially high when the price of crude oil was low. As such, the subsidies function as a buffer against the transmission of changes in the price of crude oil to the price of corn. Finally, Natanelov et al. (2011) argue that co-movement between crude oil and agricultural commodity prices should be seen as a temporal concept and thus should be assessed as such. Hereby, the authors stress the importance of “policy interventions, changing weather patterns, economic crises, changes in price interactions, geopolitics and a rising global population” in explaining the intricate price dynamics that characterise crude oil and agricultural commodity markets (Natanelov et al., 2011, p. 4892).
An alternative explanation for the “threshold co-movement” of agricultural commodity prices in and about the Financial Crisis of 2007-2008 is the simultaneous occurrence of a food, energy and financial crisis, as has been put forward by Matesanz, Torgler, Dabat and Ortega (2014). In support of their claim, the authors present the results of their network analysis, which in addition to providing further evidence of a biofuel channel, reveals that the level of co-movement spiked from mid 2008 to end 2009. Afterwards, co-co-movement levels were observed to quickly return to their historical levels. Consequently, Matesanz et al. (2014) argue that the intense uncertainty following these crises is a more plausible explanation for the temporary spike in co-movement between the crude oil and agricultural commodity prices.
Furthermore, some studies attribute (some of) the relationship between the price dynamics of crude oil and agricultural markets to aggregate demand. A good example is the study conducted by Baumeister and Kilian (2014) who, despite finding evidence for the presence of a biofuel channel, argue that the simultaneous rise in the prices of crude oil and agricultural commodities from 2003-2008 occurred against the backdrop of a booming global economy, particularly in Asia. A study conducted by Hochman et al. (2010) reaches a similar conclusion, reasoning that while biofuels and the cost-push effect contributed to the agricultural commodity price spike in 2003-2008, a rapid increase in global demand from emerging economies should be seen as the main cause. However, this line of reasoning is not without its controversy either. To illustrate, a report by the World Bank (2014) concludes that there has not been a systemic shift in the share of consumption for most agricultural commodities by large emerging economies, when comparing their share of global consumption in 2010-2012 with that in 1990-1992. An exception hereto are soybeans and corn, where in the case of China a significant increase in the share of global consumption can be observed (World Bank, 2014). This in itself is not surprising given the fact that these commodities are used to meet the rapidly increasing consumption of pork in China (Zhang & Reed, 2008). Furthermore, Baffes and Etienne (2004) find the income elasticity of most agricultural commodities to be small or close to zero, further downplaying the possibility of income growth being the common cause for the convergence of price dynamics between the oil and agricultural commodity markets.
The research of Nazlioglu and Soytas (2011) adds to the research of Zhang et al. (2010) by taking a more local perspective, although finding similar results when studying the direct and indirect effects of changes in the Brent crude oil price on the price of several of agricultural commodities in Turkey: wheat, corn, cotton, soybeans and sunflower seeds.
To put it all in a nutshell, while the existing literature is diverse of nature, so unfortunately are its findings. Nonetheless, what is clear is that most studies point towards the existence of some form of a relationship between the price dynamics of world crude oil and global agricultural markets. Herein, the majority of the literature recognises the importance of the cost-push effect as well as the biofuel channel, while the literature is more divided about the role of aggregate demand. Nonetheless, given the complexity of the price dynamics of commodity markets, the role of aggregate demand in explaining the similar price dynamics of oil and agricultural markets should not be excluded.
2.3. Hypotheses & Relevance 2.3.1. Hypotheses
Theoretical explanations such as the cost-push effect, the biofuel channel and aggregate demand, complemented by generally supportive empirical evidence for the presence of a long-run relationship between the price dynamics of world crude oil and global agricultural commodity markets, suggest a relationship between these prices. Accordingly, the main hypothesis has been formulated as following:
Hypothesis I: In the long run, the price dynamics of world crude oil markets express strong similarities with those of global agricultural commodity markets.
In respect to the origin of the relationship between the price of crude oil and those of the agricultural commodities, three supporting propositions are made. The first of these concerns the cost-push effect for which, despite the mixed empirical evidence, the theoretical underpinning is solid. In turn, the theory is strengthened by the fact that energy intensive inputs such as fuel and fertilizer take up a significant share of the production costs for commodities such as corn, wheat, soybean, rice and sugar (Sands & Westcott, 2011). In turn, for cocoa and coffee, where the use of energy intensive inputs is relatively low, the size of the cost-push effect is expected to be present but marginal (Wessel & Quist-Wessel, 2015; ICO, 2016; ICO, 2019). Based on the above, the following hypothesis has been formulated:
Hypothesis II: The cost-push effect gives rise to a long-run relationship between the global agricultural commodity prices of wheat, soybean, soybean oil, corn and rice, and the world price of crude oil. For cocoa and coffee, the cost-push effect is too small to give rise to an observable long-run relationship.
biofuels: soybeans and corn. In respect to sugar(cane) and wheat, previous empirical analysis has demonstrated mixed results. Nevertheless, the fact remains that these classify as important inputs for biofuel production, making it likely that changes in the price of crude oil impact these commodity prices as well. As such, the hypothesis below has been drafted:
Hypothesis III: The biofuel channel strengthens the long-run relationship between the global agricultural commodity prices of soybeans, corn, sugar and wheat, and the world price of crude oil over time. For the relationship between cocoa, coffee and rice, and the price of crude oil, no material changes are expected.
The last supportive proposition covers aggregate demand. Based upon the evidence presented, it appears that the explanatory power of aggregate demand is predominantly relevant for the period of 2003-2008 when the oil and agricultural commodity markets were argued to be “caught off guard” by the rapid economic growth of China, India and other emerging economies. Given the limited timespan as well as the thin empirical evidence, the following hypothesis will be put to the test:
Hypothesis IV: The impact of aggregate demand on the relationship between world crude oil and global agricultural commodity prices is limited and does not contribute to the formation of a long-run relationship between these prices.
2.3.2. Relevance
Given the importance of comprehending the intricate price dynamics between world crude oil markets and global agricultural commodity markets, in addition to the inconclusive state of the current literature, there is a clear added value to conducting further research. This paper aims to complement the existing literature by accounting for several issues that previous research has, or could not, account for.
to analyse the long-run relationship between the price dynamics of world crude oil markets and global agricultural commodity markets.
A second limitation of the existing literature is its scope. To exemplify, earlier studies have tended to focus on two extremes, taking either an overly large perspective by assessing the relationship between agricultural and oil markets through food, beverage and energy price indices or by confining their analysis to a limited selection of agricultural commodities. Consequently, whereas the former comes with a considerable loss of information as to the relationships between the price of crude oil and those of the respective agricultural commodities, the latter fails to account for the impact of changes in the world price of crude oil on agricultural commodity markets as a whole. Herewith, these studies often fail to consider the fact that agricultural markets are highly interconnected, making it likely that shocks are transmitted from one agricultural commodity price to another (Nazlioglu & Soytas, 2011). In contrast, this paper aims to provide for a more far-reaching analysis of the relationship between the price dynamics of world crude oil and global agricultural markets by including a broad selection of agricultural commodities: corn, wheat, soybean, soybean oil, sugar, rice, Arabica coffee, Robusta coffee and cocoa.
Last but not least, this study aims to contribute to the existing literature by expanding the timespan of the analysis. As the observant reader may have noted, much of the literature focuses on the period of 1990-2015 as interest was sparked by the commodity price boom in 2003-2008 (Baffes, 2011). However, from 2015 onwards the activity of scholars within this field has been moderate. Consequently, there is scant evidence on the relationship between the price dynamics of world crude oil markets and global agricultural commodity markets for more recent times. As such, by covering the period from the 2nd January of 1990 till the 18th of
3.
Data
In this study, the relationship between the price of crude oil and the prices of agricultural commodities is analysed over the period of the 2nd of January 1990 to the 18th of September
2020. Having noted that data frequency plays an important role in the ability to identify the linkages between the price dynamics of the world crude oil and global agricultural commodity markets, this study relies upon daily future price data (Zhang & Reed, 2008).1 Furthermore, the
use of futures prices allows this analysis to benefit from the important role future markets play in price discovery (Natanelov et al., 2011). Price discovery refers to the process of establishing the equilibrium price for a good through the interactions of buyers and seller via a common marketplace. Given the low transaction costs, high liquidity and centralised marketplace, the bids and offers on futures markets constantly adjust in response to new information (CME, n.d.; Yang, Li & Wang, 2020). Herewith, futures prices are not only a good representation of supply and demand conditions, but they also incorporate other influential factors such as exchange rates, economic crises and supply or demand shocks (Zafeiriou et al., 2018). Aided by high liquidity and low transaction costs, future markets often lead in the process of price discovery over cash markets (Yang et al., 2020) Consequently, futures prices lend themselves well to identify the relationships between the world crude oil and agricultural commodity prices.2
Within this study, the crude oil price data originates from the Brent crude oil contract, which is regarded as the benchmark for world oil pricing (Sheppard, McCormick, Raval, Brower & Lockett, 2020; Wang et al., 2014). Additionally, price data of the WTI crude oil contract has also been collected to allow for a complementary analysis to assure the robustness of the results. For the agricultural commodities, the price data is derived from futures contracts of cocoa, coffee, soybeans, soybean oil, wheat, rice, sugar and corn.3 The relevant futures
contracts, illustrated in footnote 2, have been selected for their high trading volumes as these serve as a good indicator of their liquidity (Kim, 2015). In turn, as high liquidity supports price discovery and market efficiency, the prices of these futures contracts provide for an accurate display of all relevant pricing information (Frijns, Indriawan & Tourani-Rad, 2018). The selection of the relevant futures contracts is line with those chosen by previous scholars (Natanelov et al., 2011; Zafeiriou et al, 2018; Zhang et al., 2010). In respect to the agricultural commodities in scope of this study, the broad selection captures each type of commodity within the agricultural markets, being softs (cocoa, sugar & coffee), grains (wheat, corn & rice) and oilseeds (soybeans & soybean oil. Furthermore, the selection of agricultural commodities allows this study to assess the relevance of the cost-push effect as it includes commodities of which the share of energy intensive inputs is low (coffee and cocoa) as well as those for which it is high (corn, soybeans, soybean oil, wheat, rice, sugar and corn). Similarly, the incorporation of commodities that are directly related to the production of biofuels (corn, soybeans, wheat and sugar) along with commodities that are not affected by it (coffee, cocoa and rice), permits this analysis to make inferences about the role of the biofuel channel. Last but not least, gold has been included in the analysis as it is an important indicator of economic conditions 1 The data originates from Investing.com, which through its partnership with Barchart Market Data solutions, sources its price data directly
from the Intercontinental Exchange (ICE) and the Chicago Mercantile Exchange (CME).
2 Further warranting the use of futures prices over spot prices is the absence of a centralised spot price for soft commodities.
3 ICE: Crude oil (Brent), Cocoa, Coffee (Arabica), Coffee (Robusta) and Sugar (#11/ World Raw). CME: Corn (No. 2 Yellow), Wheat (No.
(Natanelov et al., 2011). Herewith, its inclusion allows for the interpretation of the futures prices in a broader macro-economic context.
The period studied ranges from the 2nd of January 1990 to the 18th of September 2020.
The start and end date of the respective period are the result of data limitations as no daily futures prices were retrievable before the 2nd of January 1990 as well as the fact that at the time
of data gathering, the 18th of September 2020 was the most recent point in time.
For the analysis, the studied time period has been split up into two subsamples, ranging from the 2nd of January 2020 to the 26th of December 2003 and from the 29th of December 2003
to the 18th of December 2020. This decision is based upon two reasons. Firstly, as portrayed in
figure 13 in annex III, from 2004 onwards the world’s biofuel production rapidly accelerated. This structurally altered the dynamics of demand and supply for not only crude oil, but also for the agricultural commodities (Ciaian & Kancs, 2011). Of particular importance in this regard is the US Energy Policy Act of 2005, which was a powerful driver behind the acceleration of the world’s biofuel production from 2006 onwards (Natanelov et al., 2011). Despite the fact that the impact of the Energy Policy Act of 2005 would only materialise in 2006, futures prices can already be seen to adjust to the initial news in April/July 2004 due to the forward-looking nature of future markets (Natanelov et al., 2011). Secondly, the cut-off date distinguishes between a time period where biofuel production was of little significance and a time period where biofuel production became an integral part of agricultural production around the world. As such, cutting the captured time period at the respective date provides this study with a time period within which a relationship between the world crude oil price and the global agricultural commodity prices, if present, is predominantly explained through the presence of a cost-push effect. Alternatively, the second time period, which captures the biofuel era, offers the ability to make inferences about the impact of biofuels on the relationship between world crude oil prices and the global prices of agricultural commodities.
The fact that previous scholars have followed a comparable approach as well as the fact that a similar conclusion can be drawn from our own graphical analysis, further strengthens the choice to adopt the 26th of December 2003 as the cut-off date (Nazliogu, 2011; Zafeiriou, et
al., 2018; Tang & Xiong, 2012).4
To account for the issue of comparing diverging currencies, all commodity prices have been indexed on the price of the 2nd of January, apart from those of rice and Robusta coffee.
For these commodities, price data only goes back to 19th of January 2007 and the 14th of January
2008, respectively. As such, their prices have been indexed on the dates above.5 Additionally,
to reduce variance and to benefit the interpretation of the results, the natural log of the price indices is used.
The data refers to 5-day weeks, reflecting the fact that trading only occurs on business days. Holidays and other missing datapoints that occurred within these 5-day weeks have been linearly interpolated. It should be noted that the impact of this exercise is minimal, given the small number of missing datapoints in comparison to the overall amount of observations.
4 See annex I for the price graphs of the commodities.
5 Separate, but similarly indexed oil price series were configured for these respective commodities. However, this proved to be unnecessary
4. Preliminary analysis
As is commonly done with time series analysis, the starting point is to analyse the characteristics of the time series visually and to substantiate these statistically. Accordingly, this chapter is dedicated to analysing the price graphs of the commodities, their descriptive statistics and the correlation between them.
4.1. Price graphs
It is warranted to start off with a graphical analysis of the price graphs of the different commodities. Herein, attention is paid to the characteristic features of the price series, including but not limited to, structural breaks, trends, volatility and co-movement with the price of Brent crude oil. Analysing these characteristics does not only help with accounting for structural breaks or trends later in the empirical analysis, but it also helps to assess the validity of its results. Given the significant number of commodities captured by this analysis, as well as the added benefit of having graphs of adequate size, the decision has been made to revert the detailed analysis of the price series to annex I. Consequently, this section shall only touch upon the main results of this analysis.
Table 1: Results of the graphical analysis of the commodity price series: 1990-2004
Notes: N.A. refers to not available. No futures price data available for rice and Robusta coffee.
As shown in table 1, commodity markets were relatively stable during the period of 1990-2004, displaying little to no trends and limited volatility. Complementary, the presence of structural breaks appears confined to the price series of crude oil and Arabica.6 In contrast, the period of
2004-2020 reflects significantly more turbulent and volatile commodity markets as well as a rise in the average price level for the majority of the commodity prices. During the period of 2004-2020, several structural breaks are observed as is illustrated in table 2. These structural breaks seem to coincide with the occurrence of the Financial Crisis in 2007 and the Oil Glut in 2014 (World Bank, 2014).
6 The structural breaks are observed through visual analysis of the price graphs.
Commodity Series specification Structural break Series specification after break
Brent oil Constant 1999 Constant + trend
WTI oil Constant 1999 Constant + trend
Gold Constant N.A. Constant
Corn Constant N.A. Constant
Wheat Constant N.A. Constant
Soybeans Constant N.A. Constant
Soybean oil Constant N.A. Constant
Sugar Constant N.A. Constant
Rice N.A. N.A. N.A.
Arabica Constant 2002 Constant + trend
Robusta N.A. N.A. N.A.
Table 2: Results of the graphical analysis of the commodity price series: 2004-2020
Notes: N.A. refers to not available.
Overall, the significant differences in volatility and price level between the two periods substantiate the choice to analyse the respective periods separately. Additionally, the observation of several structural breaks warrants the use of econometric methods that account for their presence.
4.2. Descriptive statistics
Table 3 shows the descriptive statistics for the prices of crude oil, gold and agricultural commodities expressed in natural logarithms over the period 1990-2004. Each of the commodities has a mean close to zero.
Table 3: Descriptive statistics of the commodity price series: 1990-2004
Notes: Std. Dev. Refers to standard deviations. *** illustrates the rejection of the null hypothesis of normality at the 1% significance level.
The positive means are those of cocoa, Arabica, soybean oil, soybean and corn in corresponding order. In contrast, the means of Brent oil, WTI oil, gold, wheat and sugar are negative. The standard deviation, the measure of deviation from the central tendency, indicates
Commodity Series specification Structural break Series specification after break
Brent oil Constant + trend 2014 Constant
WTI oil Constant + trend 2014 Constant
Gold Constant + trend 2013 Constant
Corn Constant 2007 & 2014 Constant
Wheat Constant N.A. Constant
Soybeans Constant 2006 Constant
Soybean oil Constant 2006 Constant
Sugar Constant N.A. Constant
Rice Constant N.A. Constant
Arabica Constant + trend N.A. Constant + trend
Robusta Constant + trend N.A. Constant + trend
Cocoa Constant + trend N.A. Constant + trend
Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis Jarque-Bera Obs.
that the prices of Arabica coffee, Brent oil, WTI oil, cocoa and sugar were most volatile, respectively. Alternatively, gold was the least volatile commodity with a standard deviation of 0.178. It can also be observed by the skewness of the commodity prices, that gold and sugar have more negative price trends on average. On the other hand, the skewness values of corn, wheat and Arabica imply that these commodities more regularly experienced positive price trends. The Kurtosis values in parallel with the results of Jarque-Bera test, indicate the absence of normality in the distribution of these time series (Shahzad et al., 2018).
Complementary, in table 4 the descriptive statistics of the commodity prices over the period 2004-2020 are provided.
Table 4: Descriptive statistics of the commodity price series: 2004-2020
Notes: Std. Dev. Refers to standard deviations. ***, ** and * reflect the rejection of the null hypothesis of normality at the at 1%, 5% and 10%
level of significance, respectively
Here it is observed that within this period, each of the commodities has a higher mean than in the period of 1990-2004, confirming the rise of average price level as concluded in section 4.1. Furthermore, the minimum value of WTI oil reflects the crash of the oil price in April, where the oil price turned negative for the WTI contract (Sheppard et al., 2020). The reason that no similar movement is visible for the Brent oil price data is due to the different mechanics behind the respective futures contracts. To illustrate, whereas the Brent contract can be cash settled, the WTI contract needs to be physically settled. This means that the respective holder of the contract must either take delivery or deliver, dependent on whether having a long or short position. Given the strain on storage in the US at the time, the drop in demand due to the pandemic and the corresponding supply glut, traders were desperate to avoid delivery. This enormous pressure led to fire sales of long futures contracts, causing the price of the WTI contract to turn negative. Under the Brent contract, the option of cash settlement allowed traders to avoid having to take physical delivery, avoiding the crash that materialised under the WTI contract. (Wittels, 2020). As for the standard deviation, an increase can be observed for almost all commodities compared to the previous period, suggesting that volatility has increased. Noteworthy is the lower standard deviation for Arabica, as well as the virtually unchanged standard deviation for cocoa, when compared to the values of 1990-2004. When
Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis Jarque-Bera Obs.
assessing the skewness values, the value of WTI oil immediately stands out, again to be explained by the crash of the oil price. Also exhibiting negative values are gold, cocoa, soybean, Robusta and Brent oil, respectively. In contrast, Arabica, rice, soybean oil, corn, wheat and sugar showcase positive values, suggesting that these more regularly demonstrate positive price trends (Shahzad et al., 2018). Further, the Kurtosis values and the results of the Jarque-Bera test reveal a similar image as for the period 1990-2004.
4.3. Correlation matrix
Table 5 reports the correlation coefficients between the crude oil prices and the prices of the agricultural commodities, as well as gold. The prices of the Brent and WTI contract are almost perfectly correlated in the first period, while showing a lower correlation in the second period. The latter may be explained by the diverging price movements in respect to the recent oil crash. Nonetheless, the correlation remains high. For the period 1990-2004, the correlation coefficients indicate a small, but negative correlation between the crude oil price and the agricultural/gold commodity prices. As such, these results suggest only a weakly negative relationship between these commodity prices, weakening the explanatory power of the cost-push effect. Alternatively, the numbers for the period 2004-2020 suggest a positive correlation between the prices of crude oil and the agricultural commodities. Of these, the high positive correlation between corn, wheat, soybeans, soybean oil, rice and the price of crude oil is particularly noteworthy. These results suggest that the introduction of biofuels has indeed strengthened the relationship between these prices and the price of crude oil. In contrast, the prices of cocoa, gold and sugar appear to share little similarities with the price of crude oil. Table 5: Correlation matrix
However, it must be stressed that these correlation coefficients provide for too little information to make robust conclusions. As such, the relationship between the price dynamics of the world crude oil markets and global agricultural markets is further studied through the use of advanced econometrics.
1990-2004
2004-2020
Brent oil WTI oil Brent oil WTI oil
Brent oil 1.00 0.99 1.00 0.83 WTI oil 0.99 1.00 0.83 1.00 Gold -0.13 -0.11 0.091 -0.04 Corn -0.25 -0.22 0.75 0.58 Sugar -0.28 -0.28 0.45 0.37 Wheat -0.21 -0.19 0.70 0.55 Soybean -0.23 -0.21 0.79 0.64 Soybean oil -0.53 -0.51 0.80 0.66
Rice N.A. N.A. 0.61 0.48
Arabica -0.43 -0.41 0.53 0.42
Robusta N.A. N.A. 0.55 0.46
5. Methodology
To assess the relationship between the price dynamics of world crude oil markets and global agricultural commodity markets, as well as to make inferences about its origin(s), this paper employs unit root, cointegration and causality tests. First, the stationarity features of the different commodity prices are assessed using unit root tests. Thereafter, an Autoregressive-Distributed Lag (ARDL) model is utilised through a “Long-Run Form and Bounds test” to analyse not only the long-run relationship between the price of crude oil and the different agricultural commodity prices, but also their short-term interactions. Last but not least, the Toda-Yamamoto causality test is employed to examine the Granger causal relationships between the world crude oil price and the global agricultural commodity prices.
5.1. Unit root tests
Within time series analysis it is of key importance to apply the appropriate methodology to prevent the incorrect specification of the applied models or even worse: the use of inappropriate models. Within the selection process of the correct empirical framework, an important role is reserved for the unit root test, which indicates the level of stationarity of the time series (Wooldridge, 2016). The concept of stationary is an important determinant in the selection process of the relevant modelling technique (Shrestha & Bhatta, 2018). For time series data to be classified as stationary, the statistical properties of the time series such as its mean, variance and covariance need to be constant over the long run. In case of shocks, their effect on the properties of the stationary variable diminishes and disappears over time, allowing the time series to return to its characteristic features. By definition, time series that do not exhibit this long-term stability are considered non-stationary, i.e. have a unit root. An unit root process refers to occasion where a time series is affected by shock which fundamentally alters its dynamics. Testing for it a unit root allows for the distinction between stationary and non-stationary variables (Kočenda & Černy, 2015).
Within this study, the Augmented Dickey Fuller (ADF), the Phillips-Perron (PP) and Kwiatkowski-Phillips-Schmidt-Shin (KPSS) unit root tests are employed to assess the stationarity properties for each of the commodity price series.
The ADF and PP test have been chosen for their ability to accommodate the more complicated and dynamic structure of financial time series (Shrestha & Bhatta, 2018). Additionally, both unit root tests are based on a different approach to account for serial correlation and heteroskedasticity. To exemplify, whereas the ADF test requires the user to control for serial correlation through including lagged differences, the PP test corrects for serial correlation and heteroskedasticity issues by adjusting the t-statistics in consideration of these issues (Kočenda & Černy, 2015). Herewith, using both the ADF and PP tests contributes to the robustness of the results. Last but not least, the widespread use of these unit root tests within the existing literature substantiates their use.
To accommodate the different characteristics of the time series, the unit root tests are modified to account for an intercept or an intercept and a trend. This results in three different models. Herein, model A tests for normal stationarity, model B for level stationarity and model C for trend stationarity (Kočenda & Černy, 2015). This approach is repeated in first differences the determine the order of integration, referring to the minimum number of differences required for the time series to obtain stationarity. To illustrate, variables that are stationary without differencing are integrated to the order of zero, I(0). Alternatively, time series that become stationary in first differences are referred to as I(1).
A shortcoming of the applied unit root tests is their inability to account for structural changes, which materialise as exogenous changes to the intercept or trend. Given the permanent nature of structural changes, unit root tests tend to interpret them as shocks with a continuous effect and consequently as non-stationary. This causes the results of the unit root tests in the presence of structural breaks to be biased towards non-stationarity regardless of the stationarity of the original time series (Kočenda & Černy, 2015). To account for this limitation, this study utilises the unit root test proposed by Vogelsang and Perron (1998), which allows for an endogenously determined structural break to re-assess the stationarity properties of the time series found to be non-stationary with the regular unit root tests. The test adopts a similar hypothesis structure as employed by the ADF and PP test.
5.2. ARDL bounds tests
In this paper, an Autoregressive Distributed Lag (ARDL) model and the corresponding ARDL Long-Run Form and Bounds test, developed by Pesaran, Shin and Smith (2001), are employed to identify the short-run dynamics and the long-run relationship between the crude oil price and the respective agricultural commodity prices as visualised in Equation 5.2.1.7
𝑙𝑛𝐴𝑔𝑟𝑖 = 𝐹(𝑙𝑛𝑂𝐼𝐿) [5.2.1]
The decision to adopt the ARDL bounds test is motivated by the two major advantages it has over its counterparts. One of these is its ability to test for the existence of a relationship between the dependent variable and independent variables, irrespective of whether these are I(0), I(1) or mixed (Pesaran et al., 2001). Herewith, the approach accommodates not only the presence of variables with a different order of integration, but it also provides for an additional robustness check against the possible misspecification of stationarity by the unit root tests under section 5.1 (Türsoy, 2017; Constantinescu & Lastauskas, 2018). Secondly, the ARDL cointegration approach is able to provide insight into the short-run dynamics as well as the long-run relationship between the dependent and independent variables while accounting for the common issues associated with non-stationary variables such as endogeneity and autocorrelation (Nagawa, Wasswa & Bbaale, 2020; Shrestha & Bhatta, 2018).
However, before the appropriate model for the ARDL bounds test is presented, it is warranted to briefly touch upon the ARDL model that precedes it. The respective model is shown in Equation 5.2.2 below.
𝑙𝑛𝐴𝑔𝑟𝑖! = 𝛽"+ 𝛽#𝑡 + 1 𝛽$𝑙𝑛𝐴𝑔𝑟𝑖!%$ & $'# + 1 𝛿$ ( $'" 𝑙𝑛𝑂𝑖𝑙!%$+ 𝜀! [5.2.2]
Within the set-up of this model, the natural log of the price index of an agricultural commodity is dependent on its own lagged values as well as the (lagged) values of the independent variable
lnOil and the error term. To capture the full scope of this study, the model as well as the
subsequent models are presented in an unspecified manner. Namely, as the primary focus is to identify the impact of changes in the price of crude oil on the individual prices of the agricultural commodities, the specifics of the ARDL model differ for each agri-oil commodity pair. To exemplify, the dependent and corresponding lagged variables are dependent on the agricultural commodity of choice. As such, the name of the dependent variable lnAgri, as well as those of its lagged values, serves as a catch-all term for the natural log of the indexed price of the different agricultural commodities. The variables 𝛽" and 𝛽#𝑇 refer to the intercept and trend, respectively. Further, lnOilrefers to natural log of the indexed price of crude oil, which is either derived from the Brent crude oil or the WTI crude oil future price. Lastly, 𝜀! reflects the error term, which should be white noise (Türsoy, 2017). The lag selection of the ARDL model is determined through the automatic lag selection functionality using the Akaike Information Criterion (AIC).
Additionally, the model is run using Huber-White standard errors to account for heteroskedasticity. These standard errors are also appropriate in case of homoskedasticity (Wooldridge, 2016).
The ARDL bounds test builds upon the model displayed in Equation 5.2.2 by expressing the basic ARDL (short-run) model in first differences (∆)as well as by adding a long-term component. The ARDL bounds test model that has been applied within this paper is reflected in Equation 5.2.3. Here, the short-term component of the model, marked by the red box, captures the short-run dynamics between the price of an agricultural commodity, the price of crude oil, the corresponding lagged values and its own lagged values. Complementary, the coefficients in the blue box are used to identify the long-run relationship between the price of the agricultural commodity and the price of crude oil.
∆𝑙𝑛𝐴𝑔𝑟𝑖 = 𝛽"+ 𝛽#𝑇 + 1 𝜆$∆𝑙𝑛𝐴𝑔𝑟𝑖!%$ & $'# + 1 𝜃$ ( $'" ∆𝑙𝑛𝑂𝑖𝑙!%$+ 𝜑#𝑙𝑛𝐴𝑔𝑟𝑖!%#+ 𝜑)𝑙𝑛𝑂𝑖𝑙!%#+ 𝜈! [5.2.3]
exogenous dummy variable is reflected by the variable 𝛽)𝐷. If appropriate, more dummy variables are added in a similar fashion.
∆𝑙𝑛𝐴𝑔𝑟𝑖 = 𝛽!+ 𝛽"𝑇 + 𝛽#𝐷 + - 𝜆$∆𝑙𝑛𝐴𝑔𝑟𝑖%&$ ' $(" + - 𝜃$ ) $(!
∆𝑙𝑛𝑂𝑖𝑙%&$+ 𝜑"𝑙𝑛𝐴𝑔𝑟𝑖%&"+ 𝜑#𝑙𝑛𝑂𝑖𝑙%&"+ 𝜈%
[5.2.4]
The presence of a long-run relationship under the ARDL bounds test model is identified with the help of the following hypotheses:
H0: 𝜑#= 𝜑)= 0 (No cointegration)
H1: 𝜑#≠ 𝜑) ≠ 0 (Cointegration)
In the ARDL bounds test, cointegration is identified using a computed F-statistic, which is compared with a critical upper bound value and a critical lower bound value. If the respective statistic is higher than the critical values set out in the upper and lower bound, the null hypothesis of no cointegration can be rejected (Türsoy, 2017). If the computed F-statistic is found to be between the critical values, the test result is inconclusive. Finally, if it is below the critical value of the lower bound, the null hypothesis cannot be rejected.
After estimation of the ARDL bounds test, several diagnostic tests are done to assure the efficiency, stability and robustness of the model (Pesaran et al., 2001). These encompass the Breusch-Godfrey Lagrange multiplier (LM) test to identify serial correlation, the Jarque-Bera test to determine whether the residuals are normally distributed as well as the cumulative sum (CUSUM) and cumulative squared sum of recursive residuals (CUSUMSQ) to assess the stability of the parameters. Heteroskedasticity is already accounted for through the Huber-White standard errors.
Upon rejection of the null hypothesis, the short and long-run relationship between the price of an agricultural commodity and the price of crude oil can be further investigated through the estimation of an error correction model (ECM). The ECM is derived from the model of the ARDL bounds test by replacing the long-run component in Equation 5.2.3 with the lagged OLS residuals of the long-run model, also known as the lagged error correction term (ECT) (Nkoro & Uko, 2016). Such is reflected within the ECM as ECTt-1. The long run model from which the lagged residuals are derived as well as the equation visualising the establishment of the lagged residuals are reflected in Equation 5.2.5 and Equation 5.2.6, respectively.
𝑙𝑛𝐴𝑔𝑟𝑖! = 𝛽"+ 𝛽#𝑇 + 𝛽2𝐷+ 𝛽#𝑙𝑛𝑂𝑖𝑙!+ 𝜀! [5.2.5]
𝐸𝐶𝑇!$#= 𝜀!$# = 𝑙𝑛𝐴𝑔𝑟𝑖!$#− (𝛽"+ 𝛽#𝑇 +𝛽2𝐷+ 𝛽#𝑙𝑛𝑂𝑖𝑙!$# [5.2.6]
∆𝑙𝑛𝐴𝑔𝑟𝑖! = 𝛽"+ 𝛽#𝑇 +𝛽2𝐷 +9 𝜆%∆𝑙𝑛𝐴𝑔𝑟𝑖!$%+ 9 𝛿%ΔlnOil!$%+ 𝜑𝐸𝐶𝑇!$#+ 𝜈! ' ()# ' %)# [5.2.7]
Here the coefficients 𝜆% and 𝛿% provide for the estimated short-run relationship among the
variables, whereas 𝜑 describes the adjustment speed per period towards the long-run equilibrium after a deviation has occurred in the short-run (Shrestha & Bhatta, 2018). Given the data frequency applied within this analysis, a period refers to a single business day. For convergence towards a long-term equilibrium to be present, 𝜑 is required to be below zero and significant. If positive, the coefficient indicates there to be divergence rather than convergence. Alternatively, if zero, 𝜑 suggests there to be no adjustment, refuting the presence of a long-run relationship (Türsoy, 2017).
5.3. Toda-Yamamoto causality tests
Complementary to the ARDL bounds test, the Toda-Yamamoto causality test is employed to gain further insight into the relationship between price dynamics of world crude oil markets and global agricultural commodity markets. Namely, where the ARDL bounds test allows for the identification of long-run relationships, it does not make any inferences about their direction. Accordingly, it remains unclear whether crude oil prices affect agricultural commodity prices, vice versa or whether the relationship is bidirectional (Shrestha & Bhatta, 2018).
A common approach to overcome this shortcoming is the Granger Causality test, which estimates the direction of the relationship (Granger, 1969; Boukhelkhal & Bengana, 2018). The test is based upon the concept of causal ordering which theorises that although two variables may be correlated by chance, it is improbable that the past values of X are better suited to predict Y, in the presence of the past values of Y and all other information, if not for a causal relationship from X to Y. Likewise, this logic can be applied vice versa to establish a causal link from Y to X. However, in the presence of an omitted confounding variable, the results may still indicate a causal mechanism between X and Y despite the absence of a direct causal link. Similarly, the absence of supportive evidence of a causal relationship does not rule out its existence as too small of an effect, omitted variable bias or nonlinearity might distort the results (Stern & Enflo, 2013). As such, caution is warranted when analysing the results of Granger Causality tests.
specified in Equation 5.3.1 and 5.3.2. The null hypotheses of non-causality are expressed as: H0: 𝜃!= 𝜃"= 0 and H0: 𝜗!= 𝜗"= 0 , respectively. 𝑙𝑛𝐴𝑔𝑟𝑖#= 𝛽$+ 𝛽!𝐷 /0 𝛼!#𝑙𝑛𝐴𝑔𝑟𝑖#%&+ 0 𝛼"#𝑙𝑛𝐴𝑔𝑟𝑖#%& '!"# &()*! ) &(! 2 + /0 𝜃!#𝑙𝑛𝑂𝑖𝑙#%& ) &(! + 0 𝜃"#𝑙𝑛𝑂𝑖𝑙#%& '!"# &()*! 2 + 𝜇!# [5.3.1] 𝑙𝑛𝑂𝑖𝑙#= 𝜌$+ 𝜌!𝐷 /0 𝜗!#𝑙𝑛𝑂𝑖𝑙#%&+ 0 𝜗"#𝑙𝑛𝑂𝑖𝑙#%& '!"# &()*! ) &(! 2 + /0 𝜗!#𝑙𝑛𝐴𝑔𝑟𝑖#%& ) &(! + 0 𝜗"#𝑙𝑛𝐴𝑔𝑟𝑖#%& '!"# &()*! 2 + 𝜇"# [5.3.2]
Except for 𝑑*+,, the variables displayed within Equation 5.3.1 and 5.3.2 are the same as those under the ARDL bounds test models. 𝑑*+, represents the maximum order of integration of the
6. Empirical findings & Discussion
6.1. Unit root tests6.1.1. Standard unit root tests
To analyse the level of stationarity, ADF, PP and KPSS unit root tests have been employed. Given the scope of the analysis, the results have been diverted to annex IV. For the period 1990-2004, the findings of the unit root tests point towards the existence of a unit root for all time series, irrespective of the model’s specifications. Accordingly, these time series are concluded to be non-stationary. Upon first differencing, all time series are found to be stationary and I(1). Less uniform are the results for 2004-2020, where the time series are found to be of different orders of integration. Furthermore, for several time series the findings of the ADF and PP test contradict those of the KPSS test. To illustrate, for sugar, rice, Arabica and WTI oil, the ADF and PP test reject the null hypothesis and therewith the presence of non-stationarity. Similarly, the results of the KPSS test also reject the null hypothesis, although in favour of non-stationarity. Upon the examination of the price graphs in annex I, the series of sugar, rice and Arabica are concluded to be stationary. As for WTI oil, the conflicting results are remarkable given the strong similarities the time series has with the time series of Brent oil of which the results uniformly point towards non-stationarity. A graphical analysis of the respective time series confirms their strong similarities as well as points towards non-stationarity. Accordingly, WTI oil is concluded to be non-stationary. The contradicting results provide for a good example as to the vulnerabilities inherent to time series analysis, especially when considering the fact that the choice of model is generally determined by the results of the preceding unit root tests. The outcome of the unit root tests favours the application of an ARDL bounds test, which allows the variables to be I(0), I(1) or mixed (Pesaran et al., 2001). Herewith, the approach does not only accommodate different orders of integration, but also allows this study to sidestep the possible issues of misspecification by the unit root tests.
6.1.2. Unit root test with structural break
Table 6: Unit root test with structural break: 1990-2004
Notes: The values reflect the t-statistics of the Vogelsang and Perron unit root test (1998). Optimal lag length of the ADF test was selected
using the Schwarz Information Criterion. Model A: non-trending data with intercept break. Model B: trending data with intercept break. Model C: trending data with intercept and trend break. Break type is innovation outlier. Break point selection is minimized Dickey-Fuller t-statistic. ***, ** and * reflect statistical significance at 1%, 5% and 10% level of significance, respectively.
Table 7: Unit root test with structural break: 2004-2020
Notes: The values reflect the t-statistics of the Vogelsang and Perron unit root test (1998). Optimal lag length of the ADF test was selected
using the Schwarz Information Criterion. Model A: non-trending data with intercept break. Model B: trending data with intercept break. Model C: trending data with intercept and trend break. Break type is innovation outlier. Break point selection is minimized Dickey-Fuller t-statistic. ***, ** and * reflect statistical significance at 1%, 5% and 10% level of significance, respectively.
The findings of the Vogelsang and Perron unit root test (1998) underline the image sketched by the results of the standard unit root tests, wherein the time series were found to be of a mixed order of integration. This conclusion is robust when accounting for the presence of a single structural break. Therewith, to accommodate the different orders of integration, as well as to account for the possible threat of misspecification, the ARDL bounds test is utilised to assess the short and long-run relationship between the world price of crude oil and the global agricultural commodity prices.
Levels Model A Model B Model C
Statistic Break Statistic Break Statistic Break
Brent oil -4.28* 16/02/1999 -4.75* 16/02/1999 -4.65 01/03/1999 WTI oil -4.18 19/02/1999 -4.56 19/02/1999 -4.53 19/02/1999 Gold -2.31 13/11/1996 -2.85 13/11/1996 -3.34 20/02/2000 Corn -3.54 28/06/1996 -3.70 13/03/1998 -3.88 19/07/1996 Wheat -3.52 20/03/1996 -4.50 22/04/1997 -4.51 22/04/1997 Soybeans -2.67 16/09/1997 -3.70 17/07/1998 -3.83 17/07/1998 Soybean oil -2.85 07/05/1998 -3.67 30/11/1998 -4.21 21/01/1999 Sugar -3.79 29/12/1997 -3.86 29/12/1997 -3.95 29/12/1997 Arabica -2.98 18/07/2000 -4.49 19/04/1994 -4.63 19/04/1994 Cocoa -3.47 05/09/2001 -3.53 05/09/2001 -4.34 19/02/1999
Commodity Model A Model B Model C
Statistic Break Statistic Break Statistic Break
Brent oil (HS) -3.43 03/09/2014 -3.97 29/08/2014 -3.89 01/09/2014
Brent oil (RC) -3.07 03/09/2014 -3.34 03/09/2014 -3.54 11/03/2009
Brent oil (Rice) -3.48 03/09/2014 -3.63 03/09/2014 -3.51 01/09/2014
