Skip to main content
Top

Open Access 17-10-2023

Quantile coherency of futures prices in palm and soybean oil markets

Author: Panos Fousekis

Published in: Journal of Economics and Finance

Activate our intelligent search to find suitable subject content or patents.

search-config
loading …

Abstract

The objective of the present work is to investigate the contemporaneous price co-movement in the futures markets of soybean and palm oil. This is pursued using quantile coherency (a statistical tool that allows for both frequency- and quantile-dependent linkages between stochastic processes) and daily futures prices from 2015 to 2023. The empirical findings suggest: (a) The co-movement between palm and soybean oil prices is not very high and, at the same time, it is asymmetric; prices in the two markets are more likely to crash than to boom together. (b) The intensity of co-movement tends to increase monotonically with the time-scale considered. However, the bulk of the adjustments to shocks tend to be completed within 1 month; the differences between coherency estimates in the medium- and in the long-run are rather small. (c) Price co-movement appears to be driven by both pure (short-run) contagion as well as by fundamental-based (long-run) contagion.
Notes

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

1 Introduction

The term vegetable oil refers to any oil coming from plant sources. Vegetable oils have one of the highest international trade shares (41%) of all agricultural commodities (OECD-FAO 2023). They are used for cooking and as a basic ingredient in the food manufacturing industry (65%), as an input in the production of cosmetics, paints, varnishes, lubricants, plastics, and feed preparations for aquaculture (20%), and as a feedstock in the biodiesel industry (15%) (OECD-FAO 2023).
Although each of the vegetable oils has its own unique properties (e.g., viscosity, freezing point, and content of unsaturated fats, fatty acids, and vitamins) they are, to a large degree, fungible unless specific characteristics are desired by end-users. The potential for substitution, in turn, implies that their prices are likely to move in tandem.
The strength and the pattern of price linkages in vegetable oil markets have attracted the attention of a number of researchers. In and Inder (1997), Owen et al. (1997), and Zhou et al. (2023) relied on multivariate cointegration techniques. In and Inder (1997) and Zhou et al. (2023) concluded that prices in vegetable oil markets were well-connected in the long-run. Owen et al. (1997), in contrast, failed to reject the null hypothesis of no cointegration. Azam et al. (2020) utilized (bivariate) Continuous Wavelets; they reported evidence of co-movement that, for most price pairs, tended to be stronger in the long-run.
In and Inder (1997), Owen et al. (1997), and Zhou et al. (2023) carried out their investigations in the time-domain; they, therefore, assumed implicitly that the relationships under study were the same in small (high-frequency) and large (low- frequency) time-scales. Commodities traders, however, may operate at different time horizons and, thus, different linkages may be relevant at different frequencies (e.g., Gallegati 2012; Barunık and Kley 2019). Azam et al. (2020), allowed for frequency-dependent co-movement. They failed, however, to account for quantile-dependent co-movement; they focused on linkages around the mean of the joint distribution of prices assuming implicitly (as In and Inder 1997; Owen et al. 1997, and Zhou et al. 2023) that the mode and the strength of the relationship was invariant to the underlying market states (e.g., bullish, bearish, normal). Nevertheless, there is plenty of empirical evidence that not accounting for quantile-dependent price interrelationships may deliver rather misleading results (e.g., Tjostheim and Hufthammer 2013; Patton 2013; Barunık and Kley 2019; Ando et al. 2022).
The objective of this work is to investigate the contemporaneous relationship between futures prices in the two largest vegetable oil markets; namely, those of palm and soybean oil1. To this end, it relies on the tool of quantile coherency (Barunık and Kley 2019) which allows co-movement to be both frequency- and quantile-dependent. Quantile coherency has been employed recently to assess the linkages between cryptocurrencies (Baumohl 2019), between gold, gold mining, oil and energy sector uncertainty indices (Naeem et al. 2020), between emerging markets stocks, gold, and oil (Mensi et al. 2021), and between China’s stock and commodities markets (Wang 2023).
Relative to earlier empirical research on the topic, the present work offers a more complete characterization of the integration between the markets of soybean and palm oil by assessing the intensity and the mode of price co-movement both at different market states as well as at different time-scales2. This type of information is relevant for research economists (who are interested in market integration and efficiency), for manufacturers utilizing vegetable oils as inputs and for participants in the respective futures markets. The strength and the pattern of the contemporaneous association between the two prices are behind the soybean-palm oil spread (BOPO) that determines whether switching from one vegetable oil to the other is beneficial. At the same time, the BOPO is watched closely by traders in commodity futures markets in order to exploit profit opportunities. The empirical results from the quantile coherency analysis may enable industrialists and futures markets participants to condition their decisions on the state of the markets and the horizons they operate.
In what follows, Section 2 presents the analytical framework and Section 3 the data and the empirical model. Section 4 presents the empirical results and Section 5 offers conclusions.

2 Analytical framework

Let \(X_{1t}\) and \(X_{2t}\) (with \(t \in z\)) be two stationary stochastic processes; let also.
their respective marginal distributions \(F_{1}\) and \(F_{2}\) and a pair (\(\tau_{1}\),\(\tau_{2}\)) (with \(\tau_{i} \in (0,1))\). The cross-covariance kernel between \(X_{1t}\) and \(X_{2t}\) is
$$\gamma^{1,2} \left( {\tau_{1} ,\tau_{2} } \right) = Cov\left( {I\left( {X_{1t} \le q_{1} \left( {\tau_{1} } \right)} \right)} \right.,\,I\left( {X_{2t} \le q_{2} \left( {\tau_{2} } \right)} \right),$$
(1)
where I(A) is the indicator factor of the event A and \(q_{i} \left( {\tau_{i} } \right) = F_{i}^{ - 1} \left( {\tau_{i} } \right) = \inf \left( {q \in R:\tau_{i} < F_{i} \left( q \right)} \right)\) is the \(q_{i}\) quantile of \(X_{it}\) (for i = 1,2). The Quantile coherency between \(X_{1t}\) and \(X_{2t}\) at frequency \(\omega \in R\) and at quantiles \(\tau_{1}\) and \(\tau_{2}\) is defined as
$$\rho^{1,2} \left( {\omega ,\tau_{1} ,\tau_{2} } \right) = \frac{{f^{1,2} \left( {\omega ,\tau_{1} ,\tau_{2} } \right)}}{{\left( {f^{1,1} (\omega ,\tau_{1} ,\tau_{2} } \right)f^{2,2} \left. {\left( {\omega ,\tau_{1} ,\tau_{2} } \right)} \right)^{0.5} }},$$
(2)
where \(f^{1,2} \left( {\omega ,\tau_{1} ,\tau_{2} } \right)\) (the cross-spectral density kernel) is the translation of \(\gamma^{1,2} \left( {\tau_{1} ,\tau_{2} } \right)\) in the frequency-domain (Barunık and Kley 2019; Baumohl 2019).
Given a set of n observations, the consistent estimator of the cross-spectral density kernel is
$${\widehat{\phi }}_{n,R}^{\mathrm{1,2}}\left(\omega ,{\tau }_{1},{\tau }_{2}\right)=\frac{2\pi }{n}\sum_{s=1}^{n-1}{W}_{n}\left(\omega -\frac{2\pi s}{n}\right){\phi }_{n,R}^{\mathrm{1,2}}\left(\frac{2\pi s}{n},{\tau }_{1},{\tau }_{2}\right),$$
(3)
where \(W_{n}\) is a sequence of weight functions (Kley et al. 2016). Accordingly, the consistent estimator of quantile coherency is
$$\widehat{\rho }^{\mathrm{1,2}}\left(\omega ,{\tau }_{1},{\tau }_{2}\right)=\frac{{\widehat{\phi }}_{n,R}^{\mathrm{1,2}}\left(\omega ,{\tau }_{1},{\tau }_{2}\right)}{\left({\widehat{\phi }}_{n,R}^{\mathrm{1,1}}(\omega ,{\tau }_{1},{\tau }_{2}\right){\widehat{\phi }}_{n,R}^{\mathrm{2,2}}{\left.\left(\omega ,{\tau }_{1},{\tau }_{2}\right)\right)}^{0.5}}.$$
(4)
Insights about the structure of linkages can be obtained through a number of within- and across-frequencies tests. For example, when one wishes to verify whether quantile coherency remains the same in bearish-bearish, normal-normal, and bullish-bullish market states she (he) may test the null hypothesis \(\rho^{1,2} \left( {\omega ,\tau ,\tau } \right) = \rho^{1,2} \left( {\omega ,\tau + k,\tau + k} \right) = \rho^{1,2} \left( {\omega ,1 - \tau ,1 - \tau } \right)\), where τ is small and 0 < κ < 0.5 is an appropriately selected number; when one’s objective is to examine whether quantile coherency exhibits exchange symmetry (for example, it is the same in bull-normal and normal-bull market states) she (he) may test the null hypothesis \(\rho^{1,2} \left( {\omega ,\tau_{1} ,\tau_{2} } \right) = \rho^{1,2} \left( {\omega ,\tau_{2} ,\tau_{1} } \right)\) with \(\tau_{1} \ne \tau_{2}\) (Tjostheim and Hufthammer 2013). Alternatively, when one wishes to determine whether quantile coherency remains the same across short-, medium-, and long-term horizons she (he) may test the null hypothesis \(\rho^{1,2} \left( {\omega_{1} ,\tau_{1} ,\tau_{2} } \right) = \rho^{1,2} \left( {\omega_{2} ,\tau_{1} ,\tau_{2} } \right) = \rho^{1,2} \left( {\omega_{3} ,\tau_{1} ,\tau_{2} } \right)\), where \(\omega_{1} > \omega_{2} > \omega_{3} .\)

3 The data and the empirical model

Figure 1 presents the evolution of daily futures prices for palm and soybean oil from 1/1/2015 to 5/31/20233. Typically, soybean oil commands a price premium over palm oil for two reasons: (a) palm oil has the highest yield of oil per hectare relative to other vegetable oils and (b) soybean oil has a better taste and is richer in proteins and in omega-3 fatty acids. For most periods the two prices tended to move in the same direction4. The BOPO spread, however, exhibited considerable volatility due to the demand and supply dynamics in each market (affected by the weather, government policies such as mandates for blending ratios or intervention in foreign trade, and fluctuation in exchange rates). The narrowing (widening) of the BOPO spread directs price-sensitive users towards soybean (palm) oil.
The empirical analysis here relies on price returns5. Moreover, to avoid spurious association, the price returns have been filtered using appropriate ARMA-GARCH models as in Tjostheim and Hufthammer (2013) and Barunık and Kley (2019) (for details see Appendix Table 8). To capture the impact of different market states on the intensity and the mode of the price relationship, quantile coherency has been estimated at pairs involving the 0.05, 0.50, and 0.95 quantile levels. The short-term (high-frequency) has been set to one week (5 stock market days), the medium-term (medium-frequency) to one month (22 stock market days), and the long-term (low-frequency) to one year (250 stock market days). All these choices are consistent with the earlier works of Naeem et al. (2020) and Mensi et al. (2021). The single and joint coefficient tests have been conducted using a Wald-type statistic
$$\mathrm\Omega=\left(R\widehat C\right)^{\prime}\left(R{\widehat V}_cR^{\prime}\right)^{-1}\left(R\widehat C\right)$$
(5)
where R is the restrictions’ matrix, C is the parameters’ vector, and \(\mathop V\limits^{ \wedge } {}_{C}\) is the bootstrap estimate of their variance–covariance matrix. Under a null, Ω follows the \(\chi^{2}\) distribution with degrees of freedom equal to the number of restrictions (Patton 2013).

4 The empirical results

The empirical implementation of the consistent estimator of quantile coherency here (Eq. (4)) utilizes the Epanechnikov kernel and the mean squared error minimizing bandwidth \(b_{n} = 0.5\left( {n^{ - 0.25} } \right)\) as in Barunık and Kley (2019). Table 1 presents coherency estimates at different quantile pairs and across small-, medium-, and large-time scales6. All estimates are positive and, the large majority of them, statistically significant at the conventional levels. There is strong evidence of co-movement in the bearish-bearish market state (the (0.05, 0.05) pair) and in the normal-normal state (the (0.5, 0.5) pair). The evidence, however, of co-movement in the bullish-bullish state is much weaker; quantile coherency for the pair (0.95, 0.95) is statistically significant only at the 10% level. Given that in well-integrated markets prices should boom and crash together (Reboredo 2012), one may conclude that the degree of integration between palm and soybean oil markets is not high. This is further reinforced by the fact that all quantile coherency estimates for the bearish-bullish and the bullish-bearish states are positive and one of them (for the (95,5) pair at the high-frequency) is statistically significant at the 10% level implying positive co-movement under extreme positive returns of palm oil and extreme negative returns for soybean oil; for a high degree of market integration, a negative sign for quantile coherencies is required at quantile pairs involving strong negative returns in one market and strong positive returns in the other. The visual evidence of Table 1 suggests further that quantile coherency tends to increase with the time-scale7. Sizable differences in the estimates exist between the high- and the medium- frequencies but very small ones between the medium- and the low- frequencies. This is an indication that adjustments in the BOPO spread due to shocks in one or both markets are likely to be completed within 1 month. Also, under all three frequencies, quantile coherencies at the normal-normal state are substantially higher than those at other market states.
Table 1
Quantile coherency
Quantile pair
Frequency
1 week
1 month
12 months
Soybean oil
Palm oil
5
5
0.239
(< 0.01)
0.488
(< 0.01)
0.506
(< 0.01)
5
50
0.211
(< 0.01)
0.331
(< 0.01)
0.342
(< 0.01)
5
95
0.001
(0.978)
0.101
(0.128)
0.111
(0.111)
50
5
0.287
(< 0.01)
0.254
(< 0.01)
0.257
(< 0.01)
50
50
0.440
(< 0.01)
0.515
(< 0.01)
0.519
(< 0.01)
50
95
0.312
(< 0.01)
0.362
(< 0.01)
0.360
(< 0.01)
95
5
0.082
(0.093)
0.111
(0.121)
0.112
(0.135)
95
50
0.209
(< 0.01)
0.339
(< 0.01)
0.346
(< 0.01)
95
95
0.149
(0.076)
0.169
(0.088)
0.175
(0.086)
p-values in parentheses under the test statistics. They have been obtained using Block Bootstrap (Politis and Romano 1994) with 1500 replications
Table 2 (panel (a)) shows the results of symmetry tests with respect to the market state in the small time-scale. The null hypothesis that quantile coherencies for the pairs (5,5), (50, 50), and (95,95) are equal is rejected at the 1% level (or less). The null hypothesis, however, that quantile coherencies are equal for the bearish-bearish and bullish-bullish states is consistent with the data. Table 2 (panel (b)) shows the results of exchange symmetry tests in the small time-scale. The null hypothesis that an exchange of quantile levels of soybean and palm oil in the joint distribution of returns does not affect coherency is consistent with the data for all three pairs considered. Table 3 shows the results of symmetry tests with respect to the market state in the medium time-scale. The null hypothesis that quantile coherency is the same is now rejected not only for the normal-normal, bearish-bearish, and bullish-bullish market states but for the bearish-bearish and the bullish-bullish states as well; co-movement in the bearish-bearish market state is (at the conventional levels of significance) stronger than in the bullish-bullish one. However, the null hypothesis of exchange symmetry is again consistent with the data. Table 4 presents the results of symmetry tests with respect to the market state in the large time-scale; these are qualitatively similar to those reported in Table 3.
Table 2
Symmetry tests (frequency 1 week)
(a) Bearish-bearish, normal-normal, and bullish-bullish states symmetry
Ho: coherency is equal at the quantile pairs
Test statistics
(5,5), (50,50), and (95,95)
 − 0.191 and 0.282
(< 0.01)
0.09
(0.423)
(5,5) and (95,95)
(b) Exchange symmetry
Ho: coherency is equal at the quantile pairs
Test statistic
(5,50) and (50,5)
 − 0.075
(0.296)
(5,95) and (95,5)
 − 0.08
(0.214)
(50,95) and (95,50)
0.103
(0.178)
(a) For the three-coefficients test, the statistics are coherency for the quantile pair in the first parenthesis minus coherency for the quantile pair in the second and coherency for the quantile pair in the second minus coherency for the quantile pair in the third parenthesis. (b) p-values in parentheses under the test statistics. They have been obtained using Block Bootstrap (Politis and Romano 1994) with 1500 replications
(a) The first (second) number in parentheses under Ho is the quantile for palm (soybean) oil returns where coherency has been estimated. (b) The test statistic is coherency for the quantile pair in the first parenthesis minus coherency for the quantile pair in the second parenthesis. (c) p-values in parentheses under the test statistics. They have been obtained using Block Bootstrap (Politis and Romano 1994) with 1500 replications
Table 3
Symmetry tests (frequency 1 month)
(a) Bearish-bearish, normal-normal, and bullish-bullish states symmetry
Ho: coherency is equal at the quantile pairs
Test statistics
(5,5), (50,50), and (95,95)
 − 0.026 and 0.346
(< 0.01)
0.319
(0.015)
(5,5) and (95,95)
(b) Exchange symmetry
Ho: coherency is equal at the quantile pairs
Test statistic
(5,50) and (50,5)
0.078
(0.406)
(5,95) and (95,5)
 − 0.01
(0.911)
(50,95) and (95,50)
0.022
(0.830)
(a) For the three-coefficients test, the statistics are coherency for the quantile pair in the first parenthesis minus coherency for the quantile pair in the second and coherency for the quantile pair in the second minus coherency for the quantile pair in the third parenthesis. (b) p-values in parentheses under the test statistics. They have been obtained using Block Bootstrap (Politis and Romano 1994) with 1500 replications
(a) The first (second) number in parentheses under Ho is the quantile for palm (soybean) oil returns where coherency has been estimated. (b) The tests statistic is coherency for the quantile pair in the first parenthesis minus coherency for the quantile pair in the second parenthesis. (c) p-values in parentheses under the test statistics. They have been obtained using Block Bootstrap (Politis and Romano 1994) with 1500 replications
Table 4
Symmetry tests (frequency 1 year)
(a) Bearish-bearish, normal-normal, and bullish-bullish states symmetry
Ho: coherency is equal at the quantile pairs
Test statistics
(5,5), (50,50), and (95,95)
 − 0.012 and 0.343
(< 0.01)
0.331
(0.015)
(5,5) and (95,95)
(b) Exchange symmetry
Ho: coherency is equal at the quantile pairs
Test statistic
(5,50) and (50,5)
0.086
(0.381)
(5,95) and (95,5)
 − 0.002
(0.985)
(50,95) and (95,50)
0.014
(0.896)
(a) For the three-coefficients test, the statistics are coherency for the quantile pair in the first parenthesis minus coherency for the quantile pair in the second and coherency for the quantile pair in the second minus coherency for the quantile pair in the third parenthesis. (b) p-values in parentheses under the test statistics. They have been obtained using Block Bootstrap (Politis and Romano 1994) with 1500 replications
(a) The first (second) number in parentheses under Ho is the quantile for palm (soybean) oil returns where coherency has been estimated. (b) The tests statistic is coherency for the quantile pair in the first parenthesis minus coherency for the quantile pair in the second parenthesis. (c) p-values in parentheses under the test statistics. They have been obtained using Block Bootstrap (Politis and Romano 1994) with 1500 replications
Table 5 presents the results of the symmetry tests across high-, medium- and low- frequencies. In three cases (pairs (5,5), (5,50), and (95,50)) the null of equally is rejected at the conventional levels of significance. Given that (from Table 1) the differences between the estimates at the medium- and the high time-scale are very small, these asymmetries are driven by the increase in quantile coherency from the high- to the medium- frequency. Table 6 shows the results of the symmetry tests for the short- and the long-run. The null of equality is rejected for the quantile pairs (5,5), (5,50), (50,5), and (95,50).
Table 5
Symmetry tests (across all three frequencies)
Ho: coherency is equal across the three frequencies at the quantile pairs
Test statistics
Ho: coherency is equal across the three frequencies at the quantile pairs
Test statistics
(5,5)
 − 0.249 and − 0.018
(< 0.01)
(50,95)
 − 0.049 and 0.001
(0.467)
(5,50)
 − 0.119 and − 0.011
(0.066)
(95,5)
 − 0.029 and − 0.005
(0.852)
(5,95)
 − 0.099 and − 0.01
(0.123)
(95,50)
 − 0.13 and -0.007
(0.028)
(50,5)
0.034 and − 0.004
(0.480)
(95,95)
 − 0.02 − 0.006
(0.774)
(50,50)
 − 0.084 and − 0.004
(0.229)
  
(a) For the three-coefficients test, the statistics are coherency in 1 week minus coherency in 1 month and coherency in 1 month minus coherency in 1 year. (b) p-values in parentheses under the test statistics. They have been obtained using Block Bootstrap (Politis and Romano 1994) with 1500 replications
Table 6
Symmetry tests (short- and long-run)
Ho: coherency is equal in large and small frequency at the quantile pairs
Test statistic
Ho: coherency is equal in large and small frequency at the quantile pairs
Test statistic
(5,5)
 − 0.267
(< 0.01)
(50,95)
 − 0.049
(0.406)
(5,50)
 − 0.131
(0.02)
(95,5)
 − 0.03
(0.613)
(5,95)
 − 0.109
(0.041)
(95,50)
 − 0.138
(0.013)
(50,5)
0.03
(0.614)
(95,95)
 − 0.026
(0.674)
(50,50)
 − 0.088
(0.111)
  
(a) For the two-coefficients test, the statistics are coherency in 1 week minus coherency in 1 year. (b) p-values in parentheses under the test statistics. They have been obtained using Block Bootstrap (Politis and Romano 1994) with 1500 replications
The Financial Economics literature (e.g., Bodart and Candelon 2009; Gallegati 2012) distinguishes between two sources of co-movement in time series; namely, pure contagion and interdependence. The former refers to an increase in co-movement in the low time-scale due to investor behavior such as panic, herding, and loss of confidence; the latter (termed also as fundamental-based contagion) refers to real linkages in all market states and across all frequencies. Here, the findings of statistically significant coherency at the bearish-bearish state in the short-run and of statistically significant coherencies at the large majority of all market states across all three frequencies appear to imply that both pure and fundamental-based contagion are behind co-movement between palm and soybean oil returns. This is in line with what was reported in the work by Azam et al. (2020).

5 Conclusions

Price interrelationships among vegetable oils are important for consumers, manufactures, and traders in futures markets. This work offers an empirical analysis of contemporaneous price co-movement in the futures markets of soybean oil and palm oil using the notion of quantile coherency that allows linkages to be both quantile- and frequency- dependent.
The findings suggest that price co-movement is not very high. This is reflected not only in the levels of the estimated coherency measures but in the presence of asymmetric co-movement under different market states as well. The low degree of integration may be explained by: (a) the unique properties of each oil and the different requirements from (at least) a part of the end-users and (b) the different demand conditions in central international markets such as those in the Southeast Asia and North America. In recent years, whereas palm oil exporters of Malaysia and Indonesia strive to induce additional demand for palm oil by reducing prices in Southeast Asia, the demand of soybean oil in North America has been strong and ever-increasing due to the rise of the biodiesel and the renewable diesel industry. As matter of fact, no more than 10% of soybean oil produced in the USA has been exported; the rest has been consumed domestically (USDA-ERS 2023). The market-specific factors may be also behind the result that price co-movement tends to be higher in the normal-normal and in the bearish-bearish states relative to the bullish-bullish one; the existing differences in requirements by end-users and in individual market dynamics are likely to set limits to the ability of traders to pass very large (positive) shocks from one vegetable oil market to the other.
The weak linkages are behind the wide fluctuations of the BOPO spread which, in turn, justify hedging from palm and soybean oil manufacturers. At the same time, the BOPO volatility on the one hand makes prediction of its future direction more difficult and on the other, it promises higher profit to traders who, with better information (or simply good luck), will manage to “beat the market”. In any case, both hedgers and speculators should always bear in mind that the state of the markets and the trading horizon do have an impact on the intensity and the mode of price linkages of palm and soybean oil.
Price dynamics in palm and soybean oil markets are, to a certain extent, determined by the policy environment. The major producers and exporters of palm oil (Indonesia and Malaysia) are contemplating raising their domestic mandatory palm oil-based biodiesel blend8. At the same time, the EU (a major importer of palm oil) has recently revised, due to environmental considerations, its biofuels policy to phase out palm oil-based biodiesel by 20309. These policy changes are going to have opposite effects on palm oil prices while soybean oil demand in the US is expected to remain robust. Therefore, the relationship between soybean and palm oil prices is likely to be an important topic for empirical analysis in the next few years.
A potential limitation of the notion of quantile coherency is that it is, for the time being, suitable for bivariate stochastic processes only. Given that it yields much richer insights about co-movement relative to its alternatives, an extension to multivariate settings is highly desirable.

Declarations

Conflict of intertest/Competing interests

The author declares that has no known conflict of interest or competing interests.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://​creativecommons.​org/​licenses/​by/​4.​0/​.

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix

Appendix

Tables 7 and 8
Table 7
KPSS tests for (weak) stationarity
 
With constant only
With deterministic trend
Log-price levels
  Soybean oil
13.562
3.894
  Palm oil
11.925
2.591
Returns
  Soybean oil
0.133
0.128
  Palm oil
0.104
0.106
The 5% critical values for the model with constant only and for the model with deterministic trend are 0.463 and 0.146, respectively
Table 8
Returns filtering
Selected lags
Soybean oil
(ARMA(1,1)-GARCH(2,2))
Palm oil
(ARMA(2,1)-GARCH(1,1))
p-value
p-value
Ljung-Box test for serial correlation
Arch-LM test for conditional heteroscadasticity
Ljung-Box test for serial correlation
Arch-LM test for conditional heteroscadasticity
1
0.631
0.163
0.479
0.825
2
0.832
0.093
0.256
0.852
4
0.713
0.269
0.433
0.658
8
0.494
0.446
0.871
0.762
12
0.615
0.443
0.308
0.361
Footnotes
1
Palm and soybean oil contribute 35% and 29%, respectively, of the global vegetable oil production. The major producers and exporters of palm oil are Indonesia and Malaysia and the main importers are India and the EU. The major producers of soybean oil are Brazil, the USA, and Argentina, the major exporters are Argentina and Brazil, and the major importers are China and India (FAO 2022).
 
2
In the time-domain analysis, commonly used methods for assessing quantile-dependence are the Copulas (e.g., Reboredo 2012), the Local Gaussian Correlation (e.g., Bampinas and Panagiotidis 2017) and the Conditional Value-at-Risk regressions (e.g., Borri 2019).
 
3
Obtained from investing.com. The daily (high-frequency) data and the sample length (more than 2000 observations) are likely to work towards more reliable quantile coherency estimates and greater power of tests (e.g., Otero et al. 2022). All earlier empirical works that employed quantile coherency utilized daily data as well.
 
4
The two prices decoupled from each other in the second quarter of 2021 and in the second and third quarter of 2022. In the first case, the breakdown of the relationship occurred due to concerns over the US soybean processing industry’s capacity to meet the new demands for soybean oil from the emerging renewable diesel (RD) industry. In the second case, palm oil prices had been plummeting after a failed government-imposed three-week (April 28 to May 23) export embargo in Indonesia when soybean oil demand in the US biodiesel and RD industries remained strong (https://​www.​czapp.​com/​analyst-insights/​why-is-soybean-oil-twice-the-price-of-palm-oil/​).
 
5
The application of the KPSS test (both with a constant only and with a time trend) suggested that log-price levels contained a unit root whereas the returns, \(\ln (p_{it} /p_{it - 1} )\) (i = palm oil, soybean oil) were weakly stationary (for details see Appendix Table 7).
 
6
The estimations have been carried out using the package Quantspec (Kley 2016).
 
7
As noted by Fernadez-Macho (2018), measures of association between price returns close to 1 at the large time-scale constitute evidence of cointegration between prices. Here, all coherency estimates in the long-run are below 0.52 and different from 1 at any reasonable level of significance. Therefore, the estimates at the large time-scale also suggest that palm and soybean oil markets are not very well- integrated to each other. This is consistent with the earlier study of Owen et al. (1997).
 
Literature
go back to reference Ando T, Greenwood-Nimmo M, Shin Y (2022) Quantile connectedness: modelling tail behavior in the topology of financial networks. Manage Sci 68:2401–2431CrossRef Ando T, Greenwood-Nimmo M, Shin Y (2022) Quantile connectedness: modelling tail behavior in the topology of financial networks. Manage Sci 68:2401–2431CrossRef
go back to reference Azam A, Sarmidi T, Nor A, Zainuddin M (2020) Co-movement among World vegetable oil prices: a wavelet-based analysis. Int J Bus Soc 21:1068–1086CrossRef Azam A, Sarmidi T, Nor A, Zainuddin M (2020) Co-movement among World vegetable oil prices: a wavelet-based analysis. Int J Bus Soc 21:1068–1086CrossRef
go back to reference Bampinas G, Panagiotidis T (2017) Oil and stock markets before and after financial crises: a local Gaussian correlation approach. J Futur Mark 37:1179–1204CrossRef Bampinas G, Panagiotidis T (2017) Oil and stock markets before and after financial crises: a local Gaussian correlation approach. J Futur Mark 37:1179–1204CrossRef
go back to reference Barunık J, Kley T (2019) Quantile coherency: a general measure for dependence between cyclical economic variables. Economet J 22:131–152CrossRef Barunık J, Kley T (2019) Quantile coherency: a general measure for dependence between cyclical economic variables. Economet J 22:131–152CrossRef
go back to reference Baumohl E (2019) Are cryptocurrencies connected to forex? A quantile cross-spectral approach. Financ Res Lett 29:363–372CrossRef Baumohl E (2019) Are cryptocurrencies connected to forex? A quantile cross-spectral approach. Financ Res Lett 29:363–372CrossRef
go back to reference Bodart V, Candelon B (2009) Evidence of interdependence and contagion using a frequency domain framework. Emerg Mark Rev 10:140–150CrossRef Bodart V, Candelon B (2009) Evidence of interdependence and contagion using a frequency domain framework. Emerg Mark Rev 10:140–150CrossRef
go back to reference Borri N (2019) Conditional tail-risk in cryptocurrency markets. J Empir Financ 50:1–19CrossRef Borri N (2019) Conditional tail-risk in cryptocurrency markets. J Empir Financ 50:1–19CrossRef
go back to reference Fernandez-Macho J (2018) Time-localized wavelet multiple regression and correlation. Physica A 492:1226–1238CrossRef Fernandez-Macho J (2018) Time-localized wavelet multiple regression and correlation. Physica A 492:1226–1238CrossRef
go back to reference Gallegati M (2012) A wavelet-based approach to test for financial market contagion. Comput Stat Data Anal 56:3491–3497CrossRef Gallegati M (2012) A wavelet-based approach to test for financial market contagion. Comput Stat Data Anal 56:3491–3497CrossRef
go back to reference In F, Inder B (1997) Long-run relationships between world vegetable oil prices. Aust J Agric Econ 41:455–470CrossRef In F, Inder B (1997) Long-run relationships between world vegetable oil prices. Aust J Agric Econ 41:455–470CrossRef
go back to reference Kley T (2016) Quantile-based spectral analysis in an object-oriented framework and a reference implementation in R: the quantspec package. J Stat Softw 70:1–27CrossRef Kley T (2016) Quantile-based spectral analysis in an object-oriented framework and a reference implementation in R: the quantspec package. J Stat Softw 70:1–27CrossRef
go back to reference Kley T, Volgushev V, Dette H, Hallin M (2016) Quantile spectral processes: asymptotic analysis and inference. Bernoulli 22:1770–1807CrossRef Kley T, Volgushev V, Dette H, Hallin M (2016) Quantile spectral processes: asymptotic analysis and inference. Bernoulli 22:1770–1807CrossRef
go back to reference Otero J, Panagiotidis T, Papapanagiotou G (2022) Multivariate cointegration and temporal aggregation: some further simulation results. Comput Econ 59:59–70CrossRef Otero J, Panagiotidis T, Papapanagiotou G (2022) Multivariate cointegration and temporal aggregation: some further simulation results. Comput Econ 59:59–70CrossRef
go back to reference Owen A, Chowdhury K, Garrido J (1997) Price interrelationships in the vegetable and tropical oil markets. Appl Econ 29:119–124CrossRef Owen A, Chowdhury K, Garrido J (1997) Price interrelationships in the vegetable and tropical oil markets. Appl Econ 29:119–124CrossRef
go back to reference Patton A (2013) Copula methods for forecasting multivariate time series. Handbook of Economic Forecasting, Elsevier, North Holland, 2B: 899–960 Patton A (2013) Copula methods for forecasting multivariate time series. Handbook of Economic Forecasting, Elsevier, North Holland, 2B: 899–960
go back to reference Politis D, Romano J (1994) Limit theorems for weakly dependent Hilbert space valued random variables with applications to the stationary bootstrap. Stat Sin 4:461–476 Politis D, Romano J (1994) Limit theorems for weakly dependent Hilbert space valued random variables with applications to the stationary bootstrap. Stat Sin 4:461–476
go back to reference Reboredo J (2012) Do food and oil prices co-move? Energy Policy 49:456–467CrossRef Reboredo J (2012) Do food and oil prices co-move? Energy Policy 49:456–467CrossRef
go back to reference Tjostheim D, Hufthammer K (2013) Local Gaussian correlation. A new measure of dependence. J Econom 172:33–48CrossRef Tjostheim D, Hufthammer K (2013) Local Gaussian correlation. A new measure of dependence. J Econom 172:33–48CrossRef
Metadata
Title
Quantile coherency of futures prices in palm and soybean oil markets
Author
Panos Fousekis
Publication date
17-10-2023
Publisher
Springer US
Published in
Journal of Economics and Finance
Print ISSN: 1055-0925
Electronic ISSN: 1938-9744
DOI
https://doi.org/10.1007/s12197-023-09647-6