Measuring the synchronisation of agricultural prices: co-movement of cycles in pig and cattle prices in Brazil, Chile and Uruguay

Abstract

Simultaneous spikes in global prices of many agricultural commodities in recent years have induced an interest in quantifying the degree of synchronisation of these movements. We suggest a conceptual framework explaining why temporally varying price synchronisation may happen and propose the concordance index for the empirical measurement of the incidence, symmetry and permanence of synchronisation. We establish that the index generates insights into time series dynamics which are complementary to those obtained from cointegration analysis. We illustrate the approach with an application for the co-movement in cyclical components of pig and cattle prices in three Latin American countries. The findings reveal moderate synchronisation levels which show asymmetric instabilities.

1. Introduction

In 2007/2008 and to a somewhats lesser extent in 2011/2012, global food markets witnessed large simultaneous price rises of many agricultural commodities. The extent of this synchronous movement at the global level highlighted the fragility of the global food situation (The Economist, 2007; FAO, WFP and IFAD, 2011). Abbott (2009), among others, stresses that international food price increases brought hardship to populations who spend large shares of their budgets on staple foods.

Several publications on the threats to global crop production and food security name synchronisation in the international food system as a central issue. Mehrabi and Ramankutty (2019) show that increased synchronisation between crop production cycles destabilises global calorie production. Tigchelaar et al. (2018) conclude that global warming tends to raise the synchronisation of maize production shocks, yielding substantial effects on world markets. Anderson et al. (2019) find that climate-related globally synchronised crop failures pose substantial risks to global food security. Homer-Dixon et al. (2015) study how simultaneous shifts in socio-ecological systems might potentially interact causing an inter-systemic crisis and how such a crisis could rapidly spread at the global level.

Synchronisation of economic variables can affect the performance of agricultural and food markets. The synchronisation of prices in particular—being a fundamental determinant of manifold economic decisions—is likely to lead to price developments which move almost in parallel across countries and commodities and, thus, expose larger numbers of consumers, producers and other stakeholders to virtually identical incentives. Mitra and Josling (2009), Kim (2010), Abbott (2012) or Schüttel et al. (2011) report widespread governmental fear of the synchronisation of national and international food prices and the broad use of trade policies aimed at preventing this. FAO, WFP and IFAD, (2011) point out that whereas large countries may be able to shelter their food markets, export restrictions exacerbate price increases in international markets and aggravate food price changes and food shortages in net food-importing countries.

Given the societal relevance of the concordance of price dynamics, objective measurement and a better understanding of such synchronisation is required to enable monitoring and the development of policy responses to cushion its adverse effects. The extant agricultural economics literature on the co-movement of prices has been dominated in recent years by the analysis of price transmission and market integration (see, e.g., Hassouneh et al., 2015; Acosta, Ihle and von Cramon-Taubadel, 2019 or Kinnucan and Lloyd, 2019 for reviews). Pindyck and Rotemberg (1990) as well as Matesanz et al. (2014) use the Pearson’s correlation coefficient and network analysis to assess excess co-movement of a broad portfolio of commodity prices. Leybourne, Lloyd and Reed (1994) distinguish strong excess co-movement and weak excess co-movement employing a cointegration-based approach. Labys, Achouch and Terraza (1999) and also Byrne, Sakemoto and Xu (2020) use a dynamic factor model to analyse excess co-movement across commodity prices. Holst and von Cramon-Taubadel (2012) use regression analysis and Emmanouilides and Fousekis (2015) a copula model.

This article adds to the literature by proposing a method for transparent and precise measurement of potentially time-varying synchronisation. Our approach formalises and objectifies the visual inspection of the degree of price co-movement which has—to the best of our knowledge—not yet been addressed in the empirical literature on agricultural price analysis. Building on Pindyck and Rotemberg (1990) and Labys, Achouch and Terraza (1999), who discuss determinants of commodity price co-movement in general, we adapt the theoretical framework suggested by Gilbert (2010) for explaining potentially temporally varying occurrences of synchronised prices and establish that this method yields insights complementary to cointegration analyses. We define the synchronisation aspect of time series co-movement as the fraction of observed periods during which the time series of interest change simultaneously in the same direction and suggest measuring such synchronisation in a bivariate as well as a multivariate setting using the concordance index developed by Harding and Pagan (2002, 2006). We implement this measurement empirically and also present a rolling-window version of the index to enable assessing the stability of co-movement over subsequent sub-periods.

We illustrate the method by applying it to cycle components of pigs and cattle prices as meat-based food is currently receiving substantial media attention in the context of societal discourses on climate change. As meat production and consumption play important roles in Latin America (OECD and FAO, 2018), we focus on Brazil, Chile and Uruguay.¹

The remainder of this paper is structured as follows. Section 2 defines how synchronisation can be quantified. Section 3 presents the concordance index as the methodological tool used for measuring synchronisation in this paper and discusses differences and similarities of this concept with cointegration analysis. Section 4 gives an overview of empirical studies measuring synchronisation in several economic contexts. It includes a review of the main determinants identified by the literature which affect the synchronisation of business cycles and of financial cycles as well as a conceptual model elaborating why synchronisation of agricultural prices might happen. Section 5 illustrates the approach suggested for the case of cycles in agricultural prices in Latin America. Sections 6 and 7 summarise the main results and discuss policy implications.

2. Defining synchronisation of price series

Harding and Pagan (2006) define two or more price series as synchronised if their turning points occur at either roughly the same time or differ by intervals that are approximately constant, that is, their turning points cluster. Such an understanding of synchronisation measures the extent to which price series or selected components of them exhibit largely identical trajectories of directional change as time advances, that is, to what extent two or more time series display co-movement.

Harding and Pagan (2006) suggest measuring synchronisation by the share of periods during which all series of interest co-move, that is, the share of periods when they are simultaneously in expansion or not.² This results in three stylised cases of synchronisation.

Panel I of Figure 1 shows strong perfect positive synchronisation (SPPS): both series being simultaneously in expansion and contraction in all periods observed, i.e. |${S_{jt}}$| and |${S_{rt}} $| are identical for all t. The case of strong perfect negative synchronisation (SPNS) is shown in panel II of Figure 1: the series are in exactly the opposite phase in all periods. Strong non-synchronisation is the third possibility when no simultaneous or opposite patterns are observable, that is, both series move independently (panel III of Figure 1).

3. Measuring synchronisation

The bivariate concordance index measures the fraction of observed periods during which series are simultaneously located in the same phase of their development (Harding and Pagan, 2002, 2006). It measures the number of periods both series are moving in the same direction, that is, being simultaneously in expansion (⁠|${S_{jt}}$| = 1) or in contraction/stagnation (⁠|${S_{jt}}$| = 0).

The product of |${S_{jt}}$| and |${S_{rt}}$| denoting the directional changes of each series from period to period equals unity if both series simultaneously increase (⁠|${y_{ \bullet t}} - {y_{ \bullet t - 1}} \gt 0$|⁠). If they simultaneously decline, then the product of |$\left( {1 - {S_{jt}}} \right)$| and |$ \left( {1 - {S_{rt}}} \right)$| equals unity.⁴ Hence, if two series are in SPPS as portrayed in panel I of Figure 1 (SPNS as in panel II of Figure 1), the concordance index takes its maximum (minimum) value of unity (zero) if |${S_{jt}} = {S_{rt}}$| (⁠|$ {S_{jt}} = 1 - {S_{rt}}$|⁠) for all periods t = 1, …, T. Empirically observed synchronisation usually lies in-between these two extreme cases.

The relevant question for supply chain stakeholders is often whether prices of several agricultural commodities develop in the same direction, i.e. whether they observationally co-move, regardless of potential causal mechanisms which might drive such co-movement (Reuters, 2019; BNE, 2020). This intuitive understanding of synchronous price developments of agricultural commodities over time is the basis of equation (2). The index transforms an observed time series into a Markov chain sequence of zeros and ones as defined in equation (1) and evaluates the degree of concordance for the directional changes of these indicator variables from period to period. A main advantage of the concordance index is, therefore, the ease of understanding for non-academics, the index correspondents to an intuitive understanding by formalising the visual impression of co-movement of price series. Thus, the concordance index represents a measure which does not rely on the evaluation of invisible causal relations or on probabilistic models which are often perceived as abstract by practitioners, but it is an objective and reproducible measurement of the observable intensity and stability of co-movement.

The index defined by equations (1)–(3) does not decompose the price series of interest into structural and random components as cointegration models do (see, e.g., Fackler and Goodwin, 2001, section 4.2.5 or Meyer and von Cramon-Taubadel, 2004).^5,6 In the index, randomness does actually play no role. The association between the parametrisation of the cointegration model and the concordance index is not unique (see the online appendix for more details). The realisations of the random component in cointegration models mainly decide upon the link between them and the concordance index. The indicator variable |${S_{jt}} = f\left( {{y_{jt - 1}} - {\beta _0} - {y_{rt - 1}}, \Delta {y_{jt - 1}}, \Delta {y_{rt - 1}}, {\varepsilon _{jt}}} \right)$| is a function of the structural as well as of the unexplained components of cointegration models (see footnote 5), with |${\varepsilon _{ \bullet t}}$| tending to dominate the realisation of |${S_{jt}}$|⁠. This shows that both methods generate complementary insights into the links between the temporal trajectories of the series of interest.

4. Overview of the literature on synchronisation

The synchronisation of time series has been addressed in various economic contexts, mainly those of business cycles and financial cycles. However, there is—to the best of our knowledge—barely any discussion in the literature about the synchronisation of agricultural prices. We reviewed 37 empirical journal publications measuring the synchronisation of economic variables (for details, see the online appendix).⁷ About half of these publications—assessing time ranges from 20 to 40 years—use correlation measures. A quarter of them use the concordance index of Harding and Pagan (2002, 2006), and five studies employ structural time series models. This literature most often analyses synchronisation of price series between countries (European Union (EU) or Organisation for Economic Co-operation and Development (OECD) member states). Synchronisation is most frequently, that is, in 80 per cent of the publications, analysed in the context of business cycles. Three papers analysed stock market cycle synchronisation, and two papers analysed synchronisation of housing prices. We found only single papers assessing the synchronization of agricultural prices, of mineral commodities prices, of per capita income, and of prices in newspaper markets. The majority of papers (89 per cent) confirms synchronisation. One paper finds weak evidence for synchronisation in industrial production and strong synchronisation in stock prices.

4.1. Determinants of the synchronisation of financial markets and business cycles

Huang et al. (2016) and Imbs (2010) point out that synchronisation is a commonly existing characteristic of financial markets especially across OECD economies. Wälti (2005) provides a comprehensive review of the literature on the synchronisation of business cycles and stock markets. He assesses determinants of the synchronisation of financial markets in several industrialised countries for a period of 25 years and finds that trade, financial integration and fixed exchange rates raise synchronisation. Moreover, similar economic institutions and a common language appear to foster synchronisation while informational asymmetries reduce it. Wälti (2011) finds robust evidence for that monetary integration results in stronger synchronisation of stock markets. Nitoi and Miruna Pochea (2019) argue that market deepening is a further crucial aspect. Mobarek et al. (2016) as well as Nitoi and Miruna Pochea (2020) find that synchronisation of stock markets varies with time.

Frankel and Rose (1998), Imbs (2004) and Calderón, Chong and Stein (2007) estimate a strong and robust positive relationship between trade intensities and business cycle synchronisation. Imbs (2004) stresses that specialisation pattern in the context of global supply chains is another crucial factor. Imbs (2004, 2006) finds that integration of financial markets significantly raises the synchronisation of business cycles. Rose and Engel (2002) report a stronger synchronisation of business cycles of countries being members of a joint currency union. Jansen and Stokman (2014) find more synchronised business cycles if countries share stronger foreign direct investment relations with each other.

4.2. Determinants of the synchronisation of agricultural prices

In their influential contribution, Fackler and Goodwin (2001) comprehensively assess definitions, theoretical explanations and empirical tools for assessing the co-movement of agricultural and food prices. They emphasise that the Law of One Price (LOP) implies that ‘regional markets that are linked by trade and arbitrage will have a common, unique price’ (p. 977). A common, unique price implies that—while price levels might differ from each other due to transaction costs and currency effects—the temporal trajectories of the various levels of the single price will be identical if the LOP holds. Prices of a homogenous commodity in various regional markets would rise and shrink in the same periods; hence, their temporal developments would be synchronous. Fackler and Goodwin (2001) mention that integration of markets is often understood as the degree of co-movement of prices in various locations. Hence, assessing the synchronisation of agricultural prices can be interpreted as gaining evidence about the level and the stability of the integration of the markets of interest.

Gilbert (2010) suggests that capital asset pricing models can serve as a useful framework for conceptualising how co-movements in agricultural prices might appear. Gilbert (2010) stresses that in agriculture demand side drivers tend to be common while supply-side drivers are more likely to be idiosyncratic. The balance between the two is likely to vary over time so that synchronisation might also vary over time.

The temporal dependence of the role of these determinants for the development of price series reported by Byrne, Sakemoto and Xu (2020), Nitoi and Miruna Pochea (2020) or Gilbert and Morgan (2010) suggests that the normalised weights |${\beta _{jet}}$| and |${\gamma _{jft}}$| are time-dependent. Each of these two types of weights may therefore follow its own temporal trajectory, which can be some type of autoregressive time-series process or some discrete Hidden Markov process (Zucchini, MacDonald and Langrock, 2017).

Synchronisation between two or more series of the type SPPS as defined in equations (2) and (3) for a time window between t₁ and t₂ will take place if the partial impacts |${\beta _{jet}}$| of at least one common component |${x_e}$| will be of similar or of almost identical magnitude for all series |${y_{jt}}$| during all these periods, that is, |${\beta _{1et}} \approx \ldots \approx {\beta _{Jet}} \approx {\beta _{et}}\,\forall\,{t_1} \le t \le {t_2}$|⁠, and if the effects of the common components dominate the effects |${\gamma _{jft}}$| of the idiosyncratic determinants, that is, |$\mathop \sum \limits_{e = 1}^E |{\beta _{et}}| \approx 1$| or |$\mathop \sum \limits_{e = 1}^E |{\beta _{et}}| \gg \mathop \sum \limits_{f = 1}^F \left| {{\gamma _{jft}}} \right|, \forall\,{t_1} \le t \le {t_2}$|⁠. In contrast, the series will develop randomly, that is, in an unsynchronised fashion, if the weights |${\beta _{jet }} $| of all common determinants |${x_e}$| differ strongly from each other or are close or equal to zero, that is, if the series-specific determinants |${w_{jf}}$| dominate the series’ trajectories, that is, |$\mathop \sum \limits_{f = 1}^F \left| {{\gamma _{jft}}} \right| \approx 1$|⁠.

The time dependence of the weights is a crucial characteristic of equation (4) as it allows for phases of stronger or weaker synchronisation to follow each other. This feature has been empirically confirmed, e.g., by Imbs (2010) or Roberts and Schlenker (2013). Moreover, in section A4.2 of the online appendix, we illustrate a recent incidence of time-varying synchronisation of prices by assessing in which months during the first half of 2020—when Covid-19 severely hit international financial markets—the prices of the largest publicly listed companies of Western Europe showed the same trajectory as the common determinant |${\beta _{et}}$| caused by this disease dominated during February and March 2020 as formulated in equation (4).

The literature contains ample evidence of periods during which agricultural commodity prices developed—almost perfectly—synchronous; often referred to as commodity price booms, price spikes or price crises.⁸ One of the most recent and most influential of these phases was the increase in global agricultural prices in 2007/2008. Gilbert (2010: 403) concludes by analysing agricultural prices from 1971 up to this synchronous price boom that ‘large common food price movements’ are better explained by common factors such as demand growth, monetary expansion and exchange rate movements as well as index-based investments in agricultural futures markets (see Pindyck and Rotemberg, 1990; Labys, Achouch and Terraza, 1999) than by commodity market-specific factors.

5. Illustration: synchronisation of agricultural price cycles in South America

5.1. Quantification of synchronisation

Price time series are decomposed into several unobserved components to be able to analyse some of them separately. Commonly, such a decomposition disaggregates the observed price data into a trend, seasonality, cycle and white noise component. During the following analysis, we focus on the cyclical component for illustrative purposes. However, the concordance index can be applied to any other component as well.

Cycles are fluctuations of time series taking place in the middle run with a typical duration lasting more than 1 year (Koester, 2010). Ideally, cycles show steady expansion phases during which the levels of subsequent cycle realisations consecutively increase and which finish with turning points (peaks). Often, an expansion is followed by a contraction phase during which the level of the cycle consecutively decreases finishing with the next turning point (trough) after which a new expansion phase starts. A single cycle is usually understood as the movement from peak to peak or from trough to trough as displayed in Figure 1. We apply the filters of Kalman (1960) and Hodrick and Prescott (1997) for extracting the cycle components from the price series. We consider various filter specifications containing a deterministic linear trend with fixed drift, that is, a slope term in the long-run trend as well as a stochastic cycle and a seasonal component.

After isolating the cycles, we use the change indicator variable outlined in equation (1) and translate them into bivariate and multivariate concordance indices via equations (2) and (3). We first assess global synchronisation for all periods observed using the concordance index to provide an aggregated measure of synchronisation during all time periods T considered. As synchronisation has been found to vary with time (Mobarek et al., 2016; Nitoi and Miruna Pochea, 2020), we next assess its temporal stability by applying a moving-window version of the index to subsequent subperiods.

5.2. Assessing the stability of synchronisation and robustness checks

We choose Ω the rolling time window to be 60 months (5 years). |${\theta _t}$| denotes the index within each window.¹⁰ This version of the index shows the development of synchronisation over a moving sub-period of constant length. Thus, it allows assessing how uniformly repeated sub-periods relate to the aggregated synchronisation estimate provided by the global concordance index for the entire observation period.

The variable |${I_{jr}}$| denotes the global concordance index between cycle |${c_j} \,\, {\rm{ of \,\,series }} \,\,{y_j}$| and cycle |${c_r} \,\,{\rm{ of \,\,series }} \,\,{y_r}$|⁠, |$I_{jr}^{\,{{\theta _t}}}$| is the rolling-window index between them and |$n$| is the total number of repeated windows considered when implementing equation (6).¹¹

Next, we use F-tests to ascertain whether the rolling-window estimates differ significantly from the stable global estimates. That is, we test whether increasingly complex parametric polynomials approximating the rolling-window estimates are compatible with the stable global concordance estimate.¹²

Finally, to characterise the length of the instabilities, that is, the average deviation of the rolling-window concordance index from the global concordance index, we calculate the share of months and their mean duration during which the bivariate rolling-window index value is above and below the value of the global index.

5.3. Data

This approach has the advantage that the magnitudes of the transformed prices are measured in percentage terms in relation to the mean of each series. Table A4 (Appendix in supplementary data at ERAE online) in the online appendix summarises the descriptive statistics of the data. For each series, means and medians are almost equal, pointing to an un-skewed distribution of prices. The coefficient of variation is 0.3 for pig producer prices in Brazil and 0.2 for Chile and Uruguay, and it amounts to 0.2 for the cattle prices of all three countries, indicating almost identical degrees of relative dispersion.

6. Results

Figures 2 and 3 shows the price cycles isolated from the price indices using the Kalman filter.¹⁴ Visual inspection of Figure 2 suggests that all three countries experienced highly synchronised developments of pig price cycles from 1990 up to ca. 2003. After 2003, the Brazilian cycle shows the widest amplitude, the Uruguayan one a sustained increase and the Chilean one the most stable amplitude until about 2013.

In comparison, the cycle components of cattle prices show much less pronounced co-movement (Figure 3). The declines and increases of the cycles start in differing years; co-movement is barely discernible.

Although visual inspection gives an initial impression of the intensity of synchronisation and its temporal stability, it is an imprecise and subjective evaluation. Therefore, we formalise the synchronisation measurement using the proposed concordance indices.¹⁵

6.1. Average synchronisation

First, we evaluate average synchronisation across the entire period. The bivariate global concordance index estimates in Table 1 suggest that pairs of cycles move in the same direction until midway to three quarters of the observed time-period. The highest synchronisation is found between the cycles of pig prices of Chile and Uruguay, and the lowest is found between cattle price cycles of Chile and Uruguay. These bivariate index estimates confirm the visual impressions from Figures 2 and 3: the three pairs of cattle price cycles show lowest global synchronisation (58 per cent, 54 per cent and 50 per cent). Intra-country synchronisation (BP-BC and UP-UC) shows highest intensities of more than 60 per cent.

The global concordance index between the pig (cattle) price cycles of all three countries amounts to 30 per cent (40 per cent), indicating that all three cycles move for 68 months (130 months) in the same direction. Such co-movement gives pig (cattle) supply chain stakeholders identical information about the direction of the medium-run change of the respective national pig (cattle) prices. Hence, the co-movement of prices also synchronises stakeholder incentives and decisions related to farm-gate prices. In contrast, the synchronisation of all three cycles appears to be well below the synchronisation intensity of the bivariate indices. This indicates that all three cycles moved only in the same direction in about one-third of the entire observation period.

6.2. Temporal stability of synchronisation

Synchronisation across the entire period only gives an average assessment. We complement the assessment of the average evidence by measuring several aspects of the stability of synchronisation as outlined in Section 5.2. Synchronisation of the cycle components of meat producer prices appears to be temporally varying (see the online appendix). Several years of close co-movement of cycles—when the concordance exceeds 0.8—are followed by years of virtually absent synchronisation, i.e. when concordance falls below 0.4. These graphs reveal rolling-window estimates which point to substantially temporally varying levels of synchronisation which cluster closer or farer around the constant global estimates. The pairs of pig price cycles of all countries witnessed decreasing synchronisation during 2008, while all pairs of cattle price cycles experienced increasing or stable synchronisation intensity since 2005. Synchronisation of the cyclical components of pig and cattle prices within Brazil shows a stable pattern. In contrast, the initially almostsperfect concordance between cycle components of pig and cattle prices of Chile has been steadily diminishing to zero by 2007 and rose again steeply during the 10 successive years. The MASD-type measure defined in equation (7) translates this visual impression into precise and objective measurements of stability (Table 2). The larger the deviation between the rolling window and the global index, the more instable the synchronisation between the cycles considered. The most temporarily instable synchronisation exists between the cycles of pig and cattle price of Chile (10.5, panel VI of Figure 4) and between the cattle price cycles of Chile and Uruguay (7.6, panel VIII of Figure 4). The least instable synchronisation is found between the cycles of pig and cattle in Brazil (1.0) and between cattle prices of Brazil and Uruguay (1.9). Instability of cycle synchronisation between all three countries is higher for cattle (BC-CC-UC, 4.9) than for pig prices (BP-CP-UP, 2.5) as is also visible in Figures A3 and A4 (Appendix in supplementary data at ERAE online) in the online appendix.

To gain insight into the structure of these deviations, we assess the symmetry in terms of the share of rolling-window estimates above and below the global concordance index, as well as by assessing the permanence of synchronisation instabilities, for example, by quantifying the mean duration of the phases above and below the global estimates (Table 3). The share of months above and below the global estimates is calculated by dividing the number of rolling-window estimates above or below the global estimate by the total number of months of rolling-window estimates observed. The permanence of the instabilities is assessed by means of dividing the sum of the durations of all continuous phases of rolling-window estimates below (above) the global estimate by the number of these phases below (above).

Table 3 shows that the synchronisation of the pig price cycles of Chile and Uruguay exhibits, with 77 per cent of all observations, the highest share of rolling-window estimates above the global index estimate. The synchronisation between the pig price cycles of Brazil and Uruguay, 69 per cent, is at times stronger than the global estimate. The largest (smallest) share below the global concordance index is 66 per cent, for the synchronisation of pig and cattle price cycles inside Uruguay (23 per cent for synchronisation of pig prices between Chile and Uruguay). The trivariate rolling-window concordance index for synchronisation of pig price cycles between all three countries is above the global estimate during 79 per cent of the observed months, while for cattle it is roughly equally frequently above and below the global estimate.

The permanence of the instability is highest for the bivariate synchronisation between pig prices in Chile and Uruguay: positive deviations from the global estimate take on average 129 months. It is lowest for the synchronisation of pig and cattle cycles within Brazil having an average duration of 28 months. The most enduring negative deviations from the global estimate are found to have an average length of 131 months for cattle synchronisation between Chile and Uruguay, while the least enduring ones are again found for the synchronisation of pig and cattle price cycles within Brazil. The stability of trivariate synchronisation of pig and cattle price cycle components differs markedly. Phases of synchronisation which are stronger than average last 132 subsequent months for pig prices but only 43 months for cattle prices, while phases of synchronisation being weaker than average take 35 months for pig are almost as twice as long for cattle prices. For all F-tests on the stability of the time polynomial approximations of the non-linear rolling-window estimates, the null hypothesis of stable, time-independent synchronisation is rejected at the 5 per cent level.¹⁶

7. Discussion and conclusions

Synchronisation of economic variables has so far mainly been examined in the contexts of business cycle and financial market analysis. In agricultural economics, the topic of synchronisation has received limited attention, although it is an important aspect for the understanding of the temporal development patterns of prices of agricultural commodities. The more synchronised agricultural prices are, the greater the implied threat for food security and potential risks for producers (Mehrabi and Ramankutty, 2019; Tigchelaar et al., 2018). Homer-Dixon et al. (2015: 6) emphasise that synchronisation which is likely to lead to simultaneous crises ‘is more biophysical in origin, more inter-systemic in manifestation, more global in scope, and more rapid in development’. The higher the number of producers and consumers exposed to synchronised price signals, the larger the aggregated effects on demand and supply, which may exacerbate extreme price developments as such co-movement also synchronises incentives and decisions of stakeholders in food supply chains.

We propose the concordance index developed by Harding and Pagan (2002, 2006) as a powerful tool to assess bi- and multivariate synchronisation between agricultural price series. We suggest a range of further measurements based on this index to assess the incidence, symmetry and permanence of potential instabilities of synchronisation. We show that the concordance index generates insights into times series dynamics which are complementary to those which can be gained from cointegration analysis.

The empirical illustration of the approach demonstrates its use to assess synchronisation between single components of price time series. We focus on the cyclical component as it translates into medium-run market signals. These components are extracted from pig and cattle farm-gate prices of Brazil, Chile and Uruguay using the Kalman and Hodrick–Prescott filters. As the time series covered more than 25 years, the question of whether synchronisation is stable across the entire range of observations is crucial. We find synchronisation between pairs of pig and cattle price cycles to be of moderate levels, that is, during about 60 per cent of the observed periods, two cycles move in the same direction. Synchronisation appears to have been unstable since the 1990s. It is found to be most unstable between the pig and cattle price cycle of Chile and between the cattle price cycles of Chile and Uruguay. Synchronisation is most stable between the cycles of pig and cattle in Brazil and between the cattle price cycles of Brazil and Uruguay.

A central policy implication of our results is that governments should enhance the monitoring of agricultural and food prices. Baltussen et al. (2019) indicate that many of the existing price monitoring systems are insufficiently developed. Improved monitoring efforts would enable live monitoring and live assessment of the synchronisation of agricultural prices across countries and along supply chains. Such monitoring can serve as an early warning system for informing policy-makers and other food supply chain stakeholders to act when synchronisation intensities of agricultural and food prices rise.

The suggested formal measurement of price synchronisation would also support ex post policy evaluation. As reported by the Financial Times (2020), synchronisation measurement is able to generate objective and reproducible evidence on the success of trade policies taken to insulate domestic markets, for example, from world market developments or vis-à-vis the price developments in competing or neighbouring countries. Moreover, this approach has the potential to be a powerful information tool for assessing the success of regional trade integration efforts when the synchronisation of economic incentives is explicitly desired.

For producers, information on the extent of commodity price across regions or countries can be profitable as it allows them to engage in a kind of natural hedge (Bjornson and Carter, 1997); in the case of declining domestic prices, for example. These insights allow them to choose exporting destinations that have a negative synchronisation with prices in their domestic market (see Figure 1) so that they could profit from increasing prices by marketing there. Moreover, producers in small countries, such as Chile or Uruguay, would be better able to form their expectations about future developments of prices in their domestic market if they know to what extent their domestic price developments are synchronised with price developments in large producing or consumption countries such as Brazil. As the quantities in such large markets are much more substantial, they will tend to dominate price formation. Knowing whether domestic prices in small national markets are currently or permanently positively or negatively synchronised with those in large national markets thus greatly facilitates anticipating domestic price developments. Insights about the stability of this synchronisation allow producers and policy-makers to anticipate the endurance of such price co-movements so that they are enabled to adjust their decisions accordingly.

The measurement of synchronisation provides insights complementary to cointegration analyses. However, and similar to cointegration analysis, it does neither explain reasons for the (partially) observed co-movement, nor does it allow drawing causal conclusions. Nevertheless, objective and precise measurement of this phenomenon is the precondition for deepening its theoretical understanding, quantifying the impacts of its determinants such as assessing potential links between price cycles and business cycles and assessing the cascade of its empirical consequences in future research.

Supplementary data

Supplementary data is available at ERAE online.

Footnotes

References