Dynamic return and volatility connectedness for dominant agricultural commodity markets during the COVID-19 pandemic era

ABSTRACT This paper explores the dynamic return and volatility connectedness for the three most relevant agricultural and livestock commodity indexes (Softs, Grains and Livestock) and a media sentiment index as the Coronavirus Media Coverage Index (MCI). To that purpose, we apply the fresh time-varying parameter vector autoregression methodology during the sample period between 1 January 2020 and 30 April 2021, that is, covering the three waves of the COVID-19 pandemic crisis. Interesting results are found in this research. First, dynamic total return and volatility connectedness fluctuate over time, reaching a peak during both the first and the third waves of the global pandemic crisis. Second, in the dynamic connectedness TO the system, we observe significant differences between markets at the level of the return connectedness measure. However, in the dynamic volatility connectedness TO, there are very few differences between some elements of the system. The Coronavirus MCI appears as the less relevant receiver FROM the system, not only in terms of dynamic return connectedness but also in volatility. Finally, regarding the net dynamic total connectedness, the Coronavirus MCI shows the highest values in return and volatility, during most of the sample period analysed.


I. Introduction
Agricultural and livestock commodities are crucial resources to the social stability and food security of many countries, particularly to the progress of developing countries, and it explains why it is so important to monitor the evolution of their prices during times of instability. Besides, commodities are increasingly viewed as substitute assets by investors looking for alternative investments for diversification purposes, particularly during periods of worldwide uncertainty.
The observed increasing cross-commodity interactions have attracted the interest not only of policymakers but also of portfolio managers and investors. policymakers try to design policies that attempt to control sharp price variations of these commodities, some of them considered as basic commodities as the agricultural ones (Diebold et al., 2017). Managers and investors increasingly consider commodities as strategic securities for asset allocation, hedging, or risk management stra-tegies, and are willing to substitute traditional investment assets that are more vulnerable to shocks as equities.
Some authors argue that the financialization of commodities has played a key role in the increasing liquidity, the improving market performance, and the increasing number of investors in these markets (Silvennoinen and Thorp 2013). However, the financialization has also promoted the increased volatility transmission across commodity markets (Caporin et al. 2021). Besides, the increasing economic and policy uncertainty that arises around periods of worldwide economic distress could also explain the spillovers in returns and volatilities across markets (Aloui, Gupta, and Miller 2016). Understanding these spillovers is crucial for market participants and policymakers in their decisionmaking processes (Baruník and Křehlík 2018). While the financialization of commodity markets affects agricultural commodities to a higher degree (Caportin et al., 2021), and macroeconomic factors would affect the livestock commodities more, it is a fact that it is of interest monitoring, controlling and understanding the price evolution of these commodity markets and their interactions, particularly during turmoil periods.
As recently documented, the emergence of the pandemic caused by the SARS-CoV-2 Coronavirus, identified as COVID-19 disease by the World Health Organization (WHO) on 11 February 2020, has affected commodity markets. Goodell (2020) defines the COVID-19 pandemic as an unprecedented episode of global crisis with a destructive economic damage never seen before. This period of economic and social turbulence, comparable to the Global Financial Crisis in 2007-2008, emerges as a critical period of time, and so much research has focused on studying the effects of the COVID-19 crisis on several financial markets as, for instance, on main commodity markets of oil and gold (e.g. Bakas and Triantafyllou 2020;Salisu, Akanni, and Raheem 2020a;Salisu, Ebuh, and Usman 2020b;Corbet et al., 2020;Conlon and McGee 2020;Wang, Shao, and Kim 2020, among others).
This study tries to fill the gap in recent research by examining the impact of the SARS-CoV-2 coronavirus crisis on 10 of the most relevant agricultural and livestock commodity markets, which are concentrated on the indexes of Grains, Softs and Livestock (the Softs index includes the soft commodities of coffee, sugar, cocoa, and cotton; the Grains index includes the agricultural commodities of wheat, corn, and soybeans; and the Livestock index includes those commodities of lean hogs, live cattle, and feeder cattle). To that purpose, we examine the effects of the COVID-19 crisis on the agricultural and livestock markets from a commodity network framework. In particular, we investigate the dynamic return and volatility connectedness between the selected agricultural and livestock commodity indexes and a media-related sentiment index, by applying the recent time-varying parameter vector autoregression (TVP-VAR) methodology of Antonakakis and Gabauer (2017). This fresh approach, also used in Gabauer and Gupta (2018) and in Antonakakis, Chatziantoniou, and Gabauer (2020), among others, lets us classify the selected commodity markets into net transmitters to the system (TO) and net receivers from the system (FROM). As suggested by Antonakakis and Gabauer (2017), the TVP-VAR methodology allows us to overcome the disadvantages of the rolling window connectedness approach proposed by Diebold and Yilmaz (2012) in case of relative short time-series data, such as the period analysed in this research that covers the three waves of the COVID-19 pandemic crisis (from 1 January 2020 to 30 April 2021).
To the best of our knowledge, the most relevant contributions of this paper are twofold. First, this research explores different dynamic return and volatility connectedness measures selecting the dominant agricultural and livestock commodity markets (Grains, Softs and Livestock) and applying for the first time the TVP-VAR methodology. Second, we propose the application of the Coronavirus Media Coverage Index (MCI) in order to study the influence of the three waves of the COVID-19 pandemic crisis on these agricultural commodity markets. Thus, combining both two contributions, we can differentiate which commodity markets are net transmitters and which net receivers between the Grains, Softs and Livestock commodity markets. Moreover, for robustness, alternative volatility and sentiment index measures are employed.
Our most relevant and interesting results are the following. First, the dynamic total return and volatility connectedness measures changes over time. In addition, these measures get a peak in the first wave of the SARS-CoV-2 pandemic crunch, showing first the effect in terms of return and later in volatility, and also in the third wave. Second, in the dynamic connectedness TO the system, we observe significant differences in the level of the return connectedness reached by each agricultural commodity market, but not in the volatility one. The Coronavirus MCI would be a less relevant receiver FROM the system, so the Coronavirus MCI reveals the highest values in the net dynamic return and volatility connectedness measures, mainly during the first and third waves of the pandemic. Regarding the agricultural commodity markets, the Grains market shows the highest net dynamic connectedness, followed by the Softs market and the Livestock commodity market. Moreover, the latter market would be the most affected commodity market during the global coronavirus pandemic crisis, in concrete in the first wave.
By considering the relevance of the agricultural commodity markets studied in this paper, our results highlight that it would be necessary to implement crucial policy measures aimed at ensuring price stability of basic commodities throughout periods of economic turmoil. Besides, investors and portfolio managers should pay special attention to these commodities since they seem to have investment diversification opportunities during periods of economic distress.
The rest of the paper is organized as follows. Section 2 provides a literature review on the dynamic connectedness in agricultural commodity markets and the influence of the COVID-19 pandemic crisis. Section 3 describes the TVP-VAR methodology. Section 4 presents the data. Section 5 offers detailed information about the empirical results. Finally, Section 6 reports the main concluding remarks.

II. Literature review and theoretical framework
In this section, we set the context in which our study is framed by reviewing the prior standing literature related to the aim of this paper and by establishing the theoretical framework of the study.

Spillovers and connectedness in commodity markets
There is vast literature on the pricing, connectedness and volatility spillovers on commodity markets. The majority of the studies that include agricultural and livestock commodities focused on examining the linkages between the return and/or volatility of these commodities, and the return and/ or volatility of either crude oil prices or energy prices (e.g. Kaltalioglu and Soyas, 2011;Serra 2011;Du and McPhail 2012;Reboredo 2012;Nazlioglu, Erdem, and Soytas 2013;Koirala et al. 2015;Cabrera and Schulz 2016;Kang, McIver, and Yoon 2017;Zhang and Broadstock 2018;Yahya, Oglend, and Dahl 2019;Dahl, Oglend, and Yahya 2019;Tiwari et al. 2020;Yip et al. 2020). The evidence found is mixed, and several studies document significant relationships between oil and agricultural commodities, while so many others have shown the opposite. For instance, Du and McPhail (2012) studied volatility spillovers between crude oil and agricultural commodities and found a significant degree of volatility transmission. Similar results can be found in Koirala et al. (2015), who examined the linkages between energy and the agricultural commodities of corn and soybean. Kang, McIver, and Yoon (2017) investigated the dynamics of return and volatility spillover indices of six commodity futures markets, finding evidence of positive equicorrelation between commodity futures market returns, and of bidirectional return and volatility spillovers across commodity futures markets. Recently, Dahl, Oglend, and Yahya (2019) analysed whether there exist spillover effects among crude oil and agricultural commodities markets and found, among other results, that there exists an asymmetric and bidirectional flow of information between crude oil and agricultural commodities and that such relationship intensifies during turmoil periods. Conversely, Kaltalioglu and Soyas (2011) found an insignificant relationship between oil and agricultural commodities and food items. Cabrera and Schulz (2016) examined the linkages between energy and agricultural markets, finding no evidence of the energy as the source causing the increasing volatility in agricultural prices. In addition, there are several studies on the connectedness between agricultural commodities and stock markets (e.g. Chevallier and Ielpo 2013;Creti, Joëts, and Mignon 2013;Delatte and López, 2013;Mensi et al. 2013;Silvennoinen and Thorp 2013;Mensi, Hammoudeh, and Kang 2015;Nagayev et al. 2016;Aït-Youcef, 2019). Again, it should be noted that there is also a lack of consensus about the evidence found related to the relationship between stock and commodity markets.
Importantly, agriculture and livestock are crucial sectors of the economy, particularly during recession periods. In this line, there is a strand of the literature that empirically investigates whether counter-cyclical movements in commodity markets can be explained by the uncertainty that arises around episodes of global distress (e.g. Schwert, 1989;Hamilton and Lin 1996;Ludvigson and Ng 2009;Paye 2012;Engle, Ghysels, and Sohn 2013;Gargano and Timmermann 2014). More recently, Bakas and Triantafyllou (2018) studied the impact of macroeconomic events on the volatility of agricultural, metals and energy commodities and found that there is a positive effect on volatility derived from such kind of shocks. In the same line, Prokopczuk, Stancu, and Symeonidis (2019) analysed the relationship between economic uncertainty and the main commodity markets of agriculture, livestock, energy and metals. They found there was a strong link between economic and financial uncertainty and commodity market volatility.

Commodity markets and COVID-19 crisis
Apart from the empirical evidence on the spillovers on commodity markets, there is an increasing and recent literature about the effects of the COVID-19 crisis on a wide set of different markets and frameworks. The recent SARS-CoV-2 coronavirus pandemic crisis provides researchers with a new scenario of global distress to empirically analyse its impact. This crisis has led to the emergence of several recent studies that explore how this economic worldwide crisis has impacted on different commodity markets. 1 For instance, Bakas and Triantafyllou (2020) studied the effects of the coronavirus pandemic on commodity prices' volatility, but they focus on a general commodity index as well as two sub-indices of crude and gold. They found that the oil market responds significantly negatively to the uncertainty related to the pandemic, whereas the effect in the gold market is positive but less significant. Other authors such as Salisu, Akanni, and Raheem (2020a) found evidence of the positive relationship between commodity returns and COVID-19 fear. Besides, Salisu, Ebuh, and Usman (2020b) provided some preliminary results about the impact of the COVID-19 crisis on both oil and stock markets. Similarly, Sharif, Aloui, and Yarovaya (2020) analysed the connectedness between the coronavirus spread and the oil and stock markets. Other authors such as Corbet et al. (2020) examined potential contagion effects of the COVID-19 pandemic on gold and also on cryptocurrencies. In the same line, the study of Conlon and McGee (2020) tries to find out whether cryptocurrencies are acting as safe haven assets given the volatility in the equity markets. Wang, Shao, and Kim (2020) examined the effects derived from the pandemic on the cross-correlations between crude oil and agricultural futures markets. They found a strong correlation between oil and sugar, and such linkages became stronger under the period of COVID-19 crisis.

Connection of uncertainty and connectedness of commodity markets
The exploration of the connection between uncertainty and commodity markets is a topic that has been widely studied in recent years in the financial literature (Adekoya and Oliyide 2021). Overall, previous studies such as Liu et al. (2017) noted the negative impact of uncertainty on financial markets in general, as well as on commodity markets in particular. However, the study of the impact of uncertainty on the interdependence between commodity markets is still under-explored (Albulescu et al., 2019;Adekoya and Oliyide 2021).
In their recent work, Adekoya and Oliyide (2021) tried to fill this gap by examining how the uncertainty that arises around the COVID-19 crisis impacts the connectedness of financial and commodity markets. They employed measures related with the COVID-19 disease instead of measures of uncertainty. This line of research relies on the use of alternative variables to measure uncertainty and is related with that strand of the literature that proposes measures that capture investor's sentiment. Tetlock (2007) and Tetlock, Saar-Tsechansky, and Macskassy (2008) emphasized the potential effect and influence of media reporting on investor sentiment since information surrounding relevant news could play an important role on investor's investment decisions (Groß-Klußmann and Hautsch 2011;Renault 2017). One of the first studies on this topic is that of Barberis et al. (1998), who showed that financial markets overreact to media news even when such news are not relevant enough. To explain the effects observed on financial markets, the literature has used different mood proxies to measure investor's sentiment, namely, proxies from surveys, market data of traditional media content (Renault 2017). The research of Tetlock (2007), Kaplanski and Levy (2010) and Su, Fang, and Yin (2017), among others, relied on proxies from media reporting to analyse the effects observed on market behaviour.
In the particular context of the COVID-19 outbreak, a proxy based on media content has arisen. It is the Coronavirus MCI that measures the quantity of coronavirus-related news compared to other kinds of news. The index is a good indicator of the percentage of sources that cover the coronavirus news among all news sources and has been used in Cepoi (2020) and Haroon and Rizvi (2020) to analyse the effects of the sentiment generated by COVID-19 news on the markets. Cepoi (2020) studied the relationship between COVID-19 related news and the stock markets of six of the most affected countries by the pandemic. They found evidence of a negative link between the number of news related with the COVID-19 and the stock returns. Haroon and Rizvi (2020) focused on the volatility of equity markets and found that the panic generated by the media is positively associated with the increase in the stock market volatility.
In contrast to other papers, this study focuses on analysing the COVID-19 related news on the returns and realized volatility of some leading agricultural commodity markets. Thus, in line with previous literature, our study sets out to test the hypothesis that uncertainty tends to have a larger effect on financial markets during periods of economic turbulence. The second hypothesis would be to test whether the effect of the coronavirus news sentiment index has a different impact on the returns of the selected commodities, in comparison with the realized volatility of the commodities, due to the fact that news related to the COVID-19 pandemic could increase the volatility in the leading agricultural commodity markets. Furthermore, as a third working hypothesis, we would like to test whether the impact of the uncertainty caused by the COVID-19 pandemic on the commodity markets analysed could possibly spill over to other financial markets.
The methodological approach used in the prior literature to study the effects of macroeconomic factors on commodity markets and crosscommodity interactions is wide, including methods as GARCH models (Du and McPhail 2012;Creti, Joëts, and Mignon 2013;Silvennoinen and Thorp 2013;Cabrera et al., 2016;Nagayev et al. 2016;Kang, McIver, and Yoon 2017;Corbet et al., 2020), copulas (Delatte and Lopez 2013;Koirala et al. 2015), regression analysis (Salisu, Akanni, and Raheem 2020a), VAR models (Bakas and Triantafyllou 2020;Salisu, Ebuh, and Usman 2020b), and causality and wavelet analysis (Sharif, Aloui, and Yarovaya 2020). In this research, we empirically adopt the novel TVP-VAR methodology developed by Antonakakis and Gabauer (2017) as an extension and an improvement of the Diebold and Yilmaz (2009 connectedness approach. We apply this methodology to analyse return and volatility spillovers between shocks derived from the SARS-CoV-2 crisis, measured by the Coronavirus MCI, and three main agricultural commodity markets, Grains, Softs and Livestock. This methodology has also been used in Gabauer and Gupta (2018), Antonakakis, Chatziantoniou, and Gabauer (2020), Adekoya and Oliyide (2021), Urom et al. (2020), and Bouri et al. (2021a;, among others. Gabauer and Gupta (2018) studied the economic policy uncertainty spillovers between the U.S.A. and Japan and found that monetary policy uncertainty is the main driver. Antonakakis, Chatziantoniou, and Gabauer (2020) examined dynamic connectedness measures of the four most traded foreign exchange rates against the U.S. dollar, finding evidence of transmission effects among countries. In a more recent research, Adekoya and Oliyide (2021), Urom et al. (2020), and Bouri et al. (2021a; focused their respective studies around the pandemic crisis period and found that the spillover effects between the markets analysed (foreign exchange markets, financial markets, crude oil market, precious metals markets) increase during the turmoil period.
The results obtained could be of great importance to investors and financial analysts, particularly in times of economic crisis, such as the current SARS-CoV-2 pandemic. The sign of the effects studied, as well as their virulence, could be of great help to portfolio managers and investment planning during periods of economic downturn. In line with recent work, such as that of Adekoya and Oliyide (2021), we attempt to shed light on this issue, in the context of the COVID-19 pandemic.

III. Methodology
The approach of network connectedness comes from the methodological innovation proposed by Diebold and Yilmaz (2009. They employ network perspectives jointly with dynamic econometric models in different economic frameworks. Although this methodology has been widely used in the prior standing literature, Antonakakis and Gabauer (2017) proposed an extension and an improvement of the method of Diebold and Yilmaz (2009) through a TVP-VAR methodology. Diebold and Yilmaz (2009 developed different versions of connectedness procedures, all of them based on the forecast error variance decomposition from vector autoregressions (VAR). The VAR method introduced by Sims (1980) solves the problem of the specification of the theoretically reduced form of an underlying system of dynamic simultaneous equations. These equations describe a vector of endogenous variables as linear functions of their own values and each other's past values. However, VAR-based measures in the Diebold and Yilmaz (2009) methodology requires to arbitrarily set the rolling window size of the VAR, making the connectedness measures very sensible to the selected size. 2 Antonakakis and Gabauer (2017) extend the Diebold and Yilmaz (2009) method by proposing a TVP-VAR measure that adjusts immediately to events while avoiding the need to set the rolling window size. The TVP-VAR methodology allows the variance-covariance matrix to vary via a Kalman filter estimation with forgetting factors, which allows capturing possible changes in the underlying structure of the data. In particular, the TVP-VAR model can be written as follows:

TVP-VAR methodology
where Y t and Y tÀ 1 represent an N � 1 and an Np � 1 dimensional vectors, respectively, β t is an N � Np dimensional time-varying coefficient matrix and t is an N � 1 dimensional error disturbance vector with an N � N time-varying variance-covariance matrix, S t . The parameters β t depend on their own values β tÀ 1 and on an N � Np dimensional error matrix, v t , with an Np � Np variancecovariance matrix, R t . The time-varying coefficients obtained from the TVP-VAR model are complicated to interpret in a standard way, i.e. attending to their values alone. Instead, two general methods are used to interpret the coefficients: the generalized impulse response functions (GIRF) and the generalized forecast error variance decompositions (GFEVD) (Koop, Pesaran, and Potter 1996;Pesaran and Shin 1998). First, in order to calculate the GIRF and the GFEVD, the VAR is transformed to its vector moving average (VMA) representation following Antonakakis and Gabauer (2017) as follows: The GIRFs measure the responses of all the variables in the system due to a shock in one variable, i. In the absence of a structural model, the differences are computed through a J-step-ahead forecast, once where variable i is shocked and once where variable i is not shocked. Hence, the difference can be accounted to the shock in variable i, which can be calculated as follows: where ψ j;t J ð Þ represents the GIRFs of variable j, J represents the forecast horizon, δ j;t the selection vector is equal to 1 on the jth position and 0 otherwise, and F tÀ 1 the information set until t À 1.
In addition, the GFEVD represents the part of the variance that one variable has over the others, and can be calculated by: From the GEFVD, the total connectedness index is constructed as follows: This connectedness index allows to compute the degree at which a shock in one variable is transmitted to other variables. The main advantages of the TVP-VAR-based measure of connectedness are the followings: (i) the measure adjusts immediately to events; (ii) the size of the rolling window is adjusted automatically, something that is an improvement of the Yilmaz (2009, 2012) connectedness approach; (iii) it can also be applied to examine dynamic connectedness mea-sures for limited time-series data (Antonakakis and Gabauer 2017). 3

The TVP-VAR TO, FROM and NET connectedness indexes
The connectedness index in Eq. (12) can be computed in order to estimate different directions of the relationship between the variables in the system. First, to measure the transmission of a shock on variable i to all the other variables j in the system, we compute the 'total directional connectedness to others' (TO) as follows: Second, to calculate the transmission of a shock on variable i received from all variables j in the system, we compute the 'total directional connectedness from others' (FROM) as follows: Furthermore, it to measure the net effect, i.e. the 'net total directional connectedness' (NET), we subtract Equation (14) to Equation (13) by: The NET index can take values below, equal or above zero. A positive value of the net total directional connectedness of variable i (C g i;t > 0) would indicate that variable i influences the system more than it could be influenced by the system itself. On the contrary, a negative value of the net total directional connectedness of variable i (C g i;t < 0) would indicate that the net total directional connectedness of variable i is driven by the system. Finally, a value of 0 of the net total directional connectedness of variable i (C g i;t ¼ 0) would indicate that variable i neither has an influence nor is influenced by the system.

IV. Data
In this study, we consider daily spot prices of three agricultural commodity indexes during the period from 2 January 2020, to 30 April 2021. The selected commodity indexes are sub-indexes of the Standard & Poors Goldman Sachs Commodity Index (S&P GSCI): the S&P GSCI Softs, the S&P GSCI Grains and the S&P GSCI Livestock. 4 The first two are also sub-indexes of the S&P GSCI Agriculture index. On the one hand, the S&P GSCI Softs commodity index, is an index for traded agricultural products defined as soft commodities: coffee, sugar, cocoa, and cotton. The S&P GSCI Grains includes the agricultural commodities of wheat (Chicago & Kansas), corn, and soybeans. Finally, the S&P GSCI Livestock includes those commodities of lean hogs, live cattle, and feeder cattle. These three sub-indexes jointly represent a significant portion of the S&P GSCI commodity index, which provides investors with a reliable and publicly available benchmark for investment performance in the agricultural and livestock commodity markets (S&P Dow Jones Indices: Index Methodology).
If we observe the evolution of the spot prices of the indexes in Figure 1, we detect large drops in the spot prices (15% in Grains, 22% in Softs, and 33% in Livestock) respect to the existing prices at the beginning of the year, and which fundamentally occurs around the epicentre of the crisis, between the end of February and the end of June. This evolution shows the impact of the COVID-19 crisis on the agricultural and livestock commodity markets through the first three waves of the pandemic.
Apart from the price indexes, we include in our study the Coronavirus MCI. This index, obtained from RavenPack, is used to measure the level of media coverage about the topic of the coronavirus. 5 The Coronavirus MCI is obtained as the ratio between the number of news sources that covers the Coronavirus and the number of all the news sources. It is calculated daily and ranges from 0 to 100 (in percentage). A value of 50% means that half of the news sources within a day deal with news related with the SARS-CoV-2 crisis. This index has also been used in Cepoi (2020) and in Haroon and Rizvi (2020), who study the relationship between sentiment generated by coronavirus-related news and volatility levels in US equity markets. Our sample covers the period of the three waves of the SARS-CoV-2 coronavirus outbreak. The heart of the first wave of the COVID-19 crisis, including subsequent spikes as those in March in Europe (17 March 2020) and those in April and July in the U.S.A. (25 April 2020, and 17 July 2020). 6 The successive second wave of the coronavirus pandemic, during the second part of 2020 and the third wave of the crisis, which corresponds to the first part of 2021. This extended period will allow us to examine the different impact that the subsequent waves of the pandemic crisis have had on the commodity markets.
We compute daily returns from the daily spot prices as r t ¼ ln P t =P tÀ 1 ð Þ where r t represents the daily return obtained from the natural logarithm between daily indexes prices, and P t and P tÀ 1 represents the indexes prices at the business days t and t À 1, respectively. Table 1 displays the main descriptive statistics and unit-root tests for the return and volatility of the three agricultural and livestock commodity indexes. In terms of returns, all variables show positive and close to zero means and medians with the lowest mean value of 0.03% for Livestock and a virtually median value of 0% for Softs. The standard deviation varies from 1.18% for Softs and Grains to 1.44% for Livestock. Besides, the returns of Softs and Livestock show negative skewness, whereas those of Grains are positive. In addition, and all variables display excess kurtosis. The standard unit root (augmented Dicky-Fuller (ADF, 1979) and Phillips-Perron (PP, 1988)) and stationarity (Kwiatkowski-Phillips-Schmidt-Shin (KPSS, 1992)) tests confirm that all log-return series are stationary. In terms of volatility, we take the log difference of the volatility series to ensure stationarity. Then, all variables present positive mean and median values, and the standard deviation ranges from 2.07% for Softs to 2.44% for Grains. Again, the variables of Softs and Livestock show negative skewness, whereas Grains present positive skewness. All variables display excess kurtosis, and the unit root and stationary tests confirm that all series are stationary.

V. Empirical results
This research analyses the connectedness in the system of the most relevant agricultural commodities (Softs, Grains and Livestock), and the Coronavirus MCI. We apply the TVP-VAR approach proposed by Antonakakis and Gabauer (2017) which is useful in the context of our analysis, i.e. with short data series. We use timevarying connectedness measures based on variance decompositions from a TVP-VAR model applied to asset returns and volatilities to examine the degree to which these agricultural commodities have been affected by the three waves of the SARS-CoV-2 crisis. We estimate total, FROM, TO and net dynamic connectedness measures, during the sample period from 1 January 2020, to 30 April 2021, whose aim is to analyse the impact of the global COVID-19 pandemic crisis on the dominant agricultural commodity markets.

Preliminary results about mean return and volatility connectedness and the total dynamic connectedness
We begin our analysis by estimating a four-variable VAR model in which we include the variables of Softs, Grains, Livestock and Coronavirus MCI. The lag order, p, of the VAR model is chosen by means of a Bayes-Schwarz Information Criterion (BIC) in the reduced form model. We set the lag p = 1 and apply the VAR(1) model to approximate the models. We made a first approach to the study of the return and volatility connectedness in the system generated by the main agricultural commodities markets, also considering the Coronavirus MCI. Specifically, we examine the average values of the contribution that each agricultural commodity makes, first, TO the system and, second, FROM the system (in terms of both return and volatility), as well as the Coronavirus MCI included in the study. Thus, Figure 2 shows the mean contribution TO the system of each variable (in return and volatility) during the three waves of the SARS-CoV-2 coronavirus crisis. The most relevant transmitter (in mean) in terms of return is the Grain commodity market, being the Coronavirus MCI the less relevant transmitter TO the system explored in this study. On the other hand, Softs become the dominant transmitters (in mean), along with the Grains commodity market, in terms of volatility.
On the other hand, Figure 3 collects the mean contribution FROM the system to each variable (in terms of return and volatility). The first result we can clearly highlight is that the Coronavirus MCI shows substantially lower values than the commodity markets analysed when exploring the connectedness FROM the system studied in this paper. This phenomenon occurs not only in terms of return but also in terms of volatility. As far as the commodity markets are concerned, we observe few differences when studying the mean connectedness FROM the system in terms of return, showing the Softs, Grains and Livestock commodity markets high and similar values, with a slightly higher value for Grains. However, when analysing the mean volatility connectedness FROM the system, Livestock would be the dominant receiver, with similar values for the Softs and Grains commodity markets. Figure 4 exhibits the dynamic total connectedness in terms of return and volatility in the system explored in this paper. The most relevant result evidences that the total connectedness of the system fluctuates over time (Gabauer and Gupta 2018;; Umar, Jareño, and Escribano 2021b; among many others). It is interesting to note that in both panels of Figure 4 we observe a significant increase in the dynamic total connectedness at the heart of the first wave of the global pandemic crisis caused by SARS-CoV-2 (first part of the shaded area). Specifically, in terms of return, a spike was observed in the month of March, coinciding with the declaration of global pandemic by the World Health Organization (WHO) on 11 March 2020. In the case of dynamic total connectedness in terms of volatility, a significant peak is observed in April, approximately one month after that detected in the dynamic total return connectedness. Therefore, in both cases (return and volatility), the dynamic total connectedness in the system sharpened at the epicentre of the first wave of the global pandemic crisis, which could show a major impact on the agricultural commodity market studied in this paper. In addition, we can also observe a spike in the dynamic total connectedness in terms of volatility, coinciding with the third wave of the pandemic, i.e. during the first months of 2021, which is timidly shown in the return connectedness measure.
All the above-mentioned results can be confirmed by the results in Table 2, which show the preliminary estimates of the static connectedness in terms of returns and volatility between the leading agricultural commodity markets studied in this research and the media sentiment index (MCI).
These results are in line with previous studies, such as Silvennoinen and Thorp (2013) and Caporin et al. (2021) among others, which state that the financialization process of commodities has improved the liquidity of the market but has also increased the volatility spillover between commodity markets, mainly in periods of economic turbulence (Aloui, Gupta, and Miller 2016).

Net dynamic return and volatility connectedness: decomposition into dynamic connectedness FROM and TO
To deepen the analysis of the dynamic connectedness in the system studied in this paper, which includes the Softs, Grains and Livestock commodity markets and the Coronavirus MCI, we analysed the net dynamic connectedness, as well as its decomposition into connectedness TO ( Figure 5) and FROM ( Figure 6) in terms of both return and volatility. As commented in the previous subsection, the dominant transmitter TO the system ( Figure 5) in terms of return (Panel A) is the Grain commodity market, then Soft, then Livestock commodity markets, maintaining this order throughout the period analysed characterized by the crisis caused by COVID-19. The highest connectedness values in all three commodity markets occur during the first wave of the pandemic. In addition, during the third wave of the pandemic, Grains and Softs show similar values, increasing the gap with respect to Livestock. On the other hand, the Coronavirus MCI shows the lowest values of transmission to the system in terms of return along the sample (except for the third wave of the SARS-CoV-2 outbreak). Moreover, it is noteworthy that the coronavirus MCI shows a significant peak as a transmitter at the beginning of the period analysed, just before the heart of the global pandemic crisis.
Regarding the dynamic volatility connectedness TO (Panel B), the observed behaviour is similar to that shown by the dynamic return connectedness.  First, the estimated connectedness values are closer than in the previous analysis, considering the system studied in this paper, except for the peak shown by the Softs commodity market during May 2020. In addition, the Coronavirus MCI is the dominant transmitter in terms of volatility during the first weeks of the study period, announcing the imminent pandemic period. Then, the Softs commodity market becomes the major transmitter during the epicentre of the first wave of the pandemic, sharing dominance with Grains during the second wave of the pandemic, the latter becoming the major transmitter in the third wave of the COVID-19 crisis. The Livestock agricultural commodity market appears as the least important transmitter in the system studied in this paper, practically throughout the entire sample period, observing values much lower than the rest of the commodity markets analysed during the first wave of the pandemic. Finally, it is remarkable that the coronavirus MCI and all the commodity markets exhibit high variability at the launch of the sample period, just before the first wave of the COVID-19 crisis.
Finally, with regard to Figure 5, it should be noted that in both panels we perceive an interesting surge in the dynamic return and volatility connectedness TO the system at the centre of the COVID-19 pandemic crisis, in concrete, in March, in terms of return and delay of a few weeks in terms of volatility. Furthermore, a milder peak is observed in the third wave of the pandemic, mainly for the connectedness measure in terms of volatility. Figure 6 collects the dynamic total return (Panel A) and volatility (Panel B) connectedness FROM the system to the commodity markets and the Coronavirus MCI. In general, we find that the evolution of the three commodity markets analysed is quite similar, with connectedness measures showing very close values. Second, the Coronavirus MCI is by far the lowest receiver of dynamic connectedness FROM the system, both in terms of return and volatility. Third, while in terms of return the biggest receivers of connectedness FROM the system are the Grains and Softs commodity markets, being at a short distance the Livestock market; however, in terms of volatility, this last one (Livestock) becomes the commodity market that presents the highest value as connectedness receiver FROM the system. Below it would be the Softs commodity market and, lastly, the Grains commodity market (except for the third wave). Fourth, we return to the phenomenon previously identified, which consists of a peak in the dynamic connectedness FROM in terms of return in March 2020, which is delayed to the month of April 2020, in the case of the dynamic volatility connectedness FROM the system. Finally, it is also interesting to note that the differences between Livestock and the rest of the commodity markets widen during the third wave (in terms of return) and the first wave (in terms of volatility) of the pandemic.
Lastly, as the difference between connectedness TO and FROM in both return and volatility, Figure 7 shows the net dynamic total return and volatility connectedness of the system analysed in this paper, which includes the most relevant agricultural commodity markets (Softs, Grains and Livestock) and the Coronavirus MCI. This connectedness measure summarizes the previous results obtained in this paper. First, the Coronavirus MCI is the most important transmitter TO the system, not only in terms of return but also in terms of volatility, during different stages of the sample period analysed, but mainly in the first wave of the pandemic, and clearly during the third wave for the net dynamic return connectedness measure. Moreover, the highest values of the net dynamic return connectedness are in the first weeks of the sample, just before the first wave of the coronavirus crisis period, and, also, in March 2020, coinciding with the declaration of the SARS-CoV-2 pandemic by the WHO, showing small upturns (in both net transmitters and net receivers) during the second and third waves of the SARS-CoV-2 crisis. However, the net dynamic volatility connectedness    reaches the highest level not only at the beginning of the sample period but also in the centre of the first wave of the COVID-19 pandemic crisis, in concrete in April 2020. Second, in general, the Grains agricultural commodity market shows a positive net dynamic return and volatility connectedness, predominantly during the first wave of the coronavirus crisis. Nonetheless, the net dynamic return and volatility connectedness exhibits negative values just prior to the declaration of the SARS-CoV-2 pandemic crisis (January and February 2020), and also in some moments of the second and third waves. Third, except in some cases where the net dynamic connectedness shows occasionally positive values in the Livestock commodity market, in the rest of the cases, such a connectedness is mostly negative, both in terms of return and volatility. Moreover, the net dynamic return connectedness has a negative peak at the beginning of the  pandemic crisis epicentre, after the institutional pandemic declaration, and it shows a huge gap with respect to the other agricultural commodity markets from April 2020 to the second wave of the pandemic crisis. In terms of volatility, the worst results are found at the beginning of the analysis period (end of January and beginning of February), as well as at the end of May 2020. Since then, this commodity market exhibits a path of evolution that is just the opposite of what we see in the Softs agricultural commodity market. Fourth, the Softs commodity market has a very erratic behaviour. Thus, the net dynamic connectedness in terms of return goes from positive to negative several times throughout the sample period analysed. Furthermore, we can identify a small positive peak in the month of February and another one at the beginning of March 2020, after the declaration of a global epidemic, from which this net dynamic return connectedness starts to decrease until reaching the lowest level in April 2020. From there, the net dynamic connectedness in terms of return recovers, becoming positive until approximately the end of the sample period. As regards the net dynamic volatility connectedness, the behaviour of the Livestock market is very similar to that observed in terms of returns. Finally, we may observe a completely opposite behaviour between the net dynamic connectedness of the Coronavirus MCI and that shown by the Livestock commodity market, both in terms of return and volatility. However, the rest of the agricultural commodity markets analysed (Softs and Grains) might share evolution with the Coronavirus MCI in many moments of the sample, in both net dynamic return and volatility connectedness. 7 Again, our results support the idea that the financialization of the commodity markets has increased the volatility transmission between them because of the rise in liquidity (Silvennoinen and Thorp 2013). These results would also confirm that economic and policy instability can justify return and volatility transmissions between commodity markets (Aloui, Gupta, and Miller 2016), such as during the period characterized by the coronavirus outbreak. Moreover, in line with Caporin et al. (2021), our results would support the idea that agricultural markets on the one hand, and livestock markets on the other, could be driven by a variety of factors, the former by the process of financialization, and the latter by macroeconomic reasons. Therefore, these results would have relevant implications for managing investments in commodity markets, considering potential interactions between these markets Figure 8.

Robustness tests
In this section, we perform a series of robustness tests to provide additional results to strengthen our analysis.
We employ two alternative measures as proxies for indexes volatility, that is the 10-and 30-day historical volatility, to estimate the volatility connectedness. These measures have also been used, among others, in Umar et al. (2021c) for similar  purposes. The results obtained from re-estimating the volatility connectedness system are displayed in Tables 3 and Table 4. The results obtained show that our findings are consistent across different volatility measures, with values of the total connectedness of the Coronavirus MCI and 10-day volatility and the Coronavirus MCI and 30-day volatility of 27.9 and 24.8, respectively Figure 9.
In addition, we employ one alternative measure to capture media sentiment, the Ravenpack's Panic Index (PI). The PI is also assumed to be accurate for capturing the dynamics of media sentiment and volatility (Haroon and Rizvi 2020;Albulescu 2020;Umar et al. 2021aUmar et al. , 2021c. We re-estimate the volatility connectedness system again and replace the Coronavirus MCI with the PI. The results are displayed in Tables 5 and Table 6.
As shown in Tables 5 and Table 6, the results are more similar than our prior findings. The total connectedness of the PI and 10-day volatility and the PI and 30-day volatility, obtained from our TVP-VAR model, are 27.6 and 27.9, respectively. As observed, overall, our results stay same. Consistently, the alternative volatility measures and media sentiment proxy give us qualitatively similar results and are robust. 8

VI. Concluding remarks
This paper examines the dynamic return and volatility connectedness in the system consisting of the three main agricultural and livestock commodity indexes (Softs, Grains and Livestock -that comprises the 10 dominant agricultural and livestock commodity markets of coffee, sugar, cocoa, cotton, wheat, corn, soybeans, lean hogs, live cattle, and feeder cattle) and the Coronavirus MCI extracted from RavenPack. In the context of an extension of the TVP-VAR connectedness methodology proposed by Antonakakis and Gabauer (2017), this study explores the net dynamic connectedness, distinguishing between TO and FROM connectedness, in both return and volatility. In addition, this methodology allows us to classify industrial metals as net transmitters TO and net receivers FROM the system.
About the sample period, this paper focuses on analysing the impact of the COVID-19 pandemic crisis, which is having repercussions worldwide comparable to the global financial crisis caused by subprime mortgages. In particular, to analyse the effect of this global pandemic outbreak on the most relevant agricultural commodity markets, our study uses the previously mentioned Coronavirus MCI, which has been applied in some very recent papers. In concrete, our sample period ranges from 1 January 2020, to 30 April 2021, in order to conduct a thorough examination of the three waves of the SARS-CoV-2 pandemic.
Fresh and interesting results for the agricultural commodity markets are reached. First, the dynamic total return and volatility connectedness fluctuate over time, reaching a peak during the heart of the first wave of the global pandemic crisis, primarily in the dynamic total return connectedness (March 2020) and later in the volatility one (April 2020), but also during the third wave of the pandemic, mainly in terms of volatility. This result would confirm the impact of the COVID-19 global pandemic on the agricultural commodity markets,  both in terms of dynamic total return and volatility connectedness, confirming the first hypothesis assumed in this research. As previously mentioned, we apply a recent methodology to split the net dynamic connectedness into connectedness FROM and TO (the system), as well as in terms of return and volatility. As far as the dynamic connectedness TO the system is concerned, we find a similar behaviour but on different levels between the agricultural commodity markets in the case of return connectedness TO. Specifically, the dominant transmitter TO the system is the Grains commodity market, then the Softs market and, lastly, the Livestock market. The Coronavirus MCI exhibits the least values of the dynamic return connectedness TO the system. However, in terms of volatility, the gap between agricultural commodity markets is lower (mainly during the second and third waves of the pandemic), stressing the Coronavirus MCI (at the beginning of the sample) and the Softs commodity market (in April-May 2020, first wave) as some of the major transmitters TO the system. On the other hand, the Livestock commodity market shows the lowest dynamic volatility connectedness primarily during the second wave of the pandemic.
With regard to the dynamic connectedness FROM, the Coronavirus MCI appears as the less relevant receiver FROM the system not only in terms of dynamic return connectedness but also in volatility. In the first case, the evolution of the return connectedness between the agricultural commodity markets is quite similar to that observed in the dynamic return connectedness TO, although the peak observed in March 2020 is higher for the Softs commodity market. In addition, differences between commodity markets are very small, although they increase during the third wave of the pandemic. In terms of volatility, the Livestock commodity market would appear as the dominant receiver FROM the system, being the least relevant receiver of the grain commodity market (except for the third wave of the pandemic). It is interesting to note that the peaks observed in the dynamic return connectedness FROM in the second half of March and in the volatility one in the middle of April 2020, as well as a slight rebound in the third wave of the COVID-19 pandemic. Apparently, the second hypothesis proposed in this paper would not be confirmed, as we did not find significant differences between the impact of coronavirus news on the connectedness in terms of return and in terms of volatility.
Lastly, when considering the net effect of the above two connectedness measures (net dynamic total connectedness), the Coronavirus MCI emerges as the most valuable net transmitter to the system both in terms of return and volatility, mainly before the first wave of the pandemic and also during the third wave. Furthermore, the agricultural commodity markets studied in this paper show the following order: The Grains commodity market, then the Softs market and finally the Livestock market, mostly for the net dynamic return connectedness measure. The Livestock commodity market exhibits negative values in both measures during the full sample period. Moreover, this commodity market presents negative peaks in terms of return to the introduction of the pandemic's heart and in terms of volatility in the centre of the SARS-CoV-2 outbreak crisis. Moreover, the gap between this commodity market and the rest of market widens mainly during the first wave of the pandemic. Finally, the third hypothesis proposed in this paper would be confirmed, as the results obtained would support the idea that contagion (connectedness) between the commodity markets analysed increases in periods of economic turbulences such as the COVID-19 crisis.
Our results have important policy implications due to the worldwide relevance of the agricultural commodity markets. The Livestock agricultural commodity market seems to be the most affected market because of the COVID-19 pandemic crisis (mainly in the first wave of the pandemic), showing a high persistence not only in return but also in volatility. As agricultural and livestock commodity markets are essential markets, it is necessary to adopt important policy actions aimed at guaranteeing price stability during periods of economic turbulence. Our results are also of interest for both portfolios managers and investors, seeking for investment opportunities through commodity markets or trying to diversify the risk of their portfolios and other hedging strategies for their investments.
As regards future lines of research, a first extension of the work would be to expand the study to other agricultural commodity markets, as well as to apply this recent TVP-VAR connectedness methodology to alternative commodity and cryptocurrency markets, in the context of the global SARS-CoV-2 pandemic crisis. Another interesting research would consist in exploring the impact of the Coronavirus MCI on the most relevant international sector indices in terms of return and volatility.
On the other hand, it would be really interesting to explore asymmetric volatility spillovers between agricultural commodity markets using high frequency tick data. The most common approach used to study the asymmetric volatility spillovers is based on Barndorff-Nielsen and Shephard (2002) and Barndorff-Nielsen, Kinnebrock, and Shephard (2008). The approach boils down to computing realized semi variances to analyse the impact of asymmetric volatility spillover by distinguishing between good and bad volatility in conjunction with the connectedness framework of Diebold and Yilmaz (see, for instance, Barunik et al, 2016;He, Wang, and Yin 2020).

Disclosure statement
No potential conflict of interest was reported by the author(s).