Short-Term Speculation Effects on Agricultural Commodity Returns and Volatility in the European Market Prior to and during the Pandemic

Abstract

Motivated by increased agricultural commodity price volatility and surges during the past decade, we investigated whether financial speculation is to blame. The aim of this paper is to build on prior research about to what extent and in which ways financial speculation undermines agricultural commodity prices. In our analysis, we utilized the daily returns on milling wheat, corn, and soybean futures from the Euronext Commodities Paris market (MATIF) as well as the short-term speculation index. To quantify this impact, we apply Granger noncausality tests as well as the GARCH (generalized autoregressive conditional heteroskedasticity) technique. We also propose a model using seasonal dummy variables to examine whether financial speculation has a greater impact on price volatility during more volatile months. According to our results, financial speculation, as an external factor, in most cases has no effect or reduces the volatility of the underlying futures prices. The opposite is observed in the corn market, where volatility has risen in the post-2020 period and has been pushed up even more by speculation in April. However, since the influence on other commodities is limited or nonexistent, more emphasis should be focused on speculation in the European corn futures market or its interdependence with energy markets.

1. Introduction

Speculative activity is common among many markets, especially those where transactional costs are minimal and goods traded are standardized and liquid, such as financial markets, including agricultural futures markets. Futures contracts are standardized agreements between two parties to acquire or sell a standardized asset of a certain quantity and quality at a fixed price at a future date. Agricultural commodity producers and consumers, also known as “commercial market participants,” employ them to protect themselves against price movements and volatility. Typically, there is a different amount of demand to hedge against increasing or decreasing prices. This results in a difference between commercial long and commercial short positions, and therefore creates risk premium opportunities, also known as hedging pressure, because commercial hedgers are frequently net short [1]. Futures market speculators seek to earn these risk premiums. In other words, they take over this price risk in exchange for earning profits. Speculators can also correct price drifts from their fundamental values, as explained by supply and demand factors. For example, according to Du and Dong [2], who investigated US dairy futures markets, the volatility of both price and trade volume can be explained by flows of new market information. Consequently, it may be argued that some speculative activity is both common and necessary in these markets to make them more efficient and liquid. However, the number of speculators in major international commodity markets has increased considerably in the last two decades because of the market liberalization and financialization of many agricultural commodities, as well as the ability to make cash settlements. However, in some cases, speculative activity can get out of hand when the number of positions held by speculators exceeds the number of positions held by commercial long and short positions. In many cases, even traditional commercial participants engage in speculative activities [3].

Traditional economic theories that investigate the role and impact of speculation on the price of assets on markets are effective market theories and behavioral finance. Even though speculators bring liquidity and new information, they have, on the other hand, different objectives than typical business users seeking to protect themselves from price risk, and their behavior patterns may result in commodity prices that are not representative of their genuine worth, therefore creating opportunities for price booms and spikes. Recent empirical work on commodity futures has extensively explored futures market volatility and what factors cause it. The cost-of-storage model, which examines inventory quantities, interest rates, and desired profitability to explain price volatility and differences between spot and futures prices, is often used to investigate speculators’ participation in futures markets [4]. In many studies, it is found that speculation in derivatives markets has no or limited statistically significant effect on price or return volatility and instead benefits the stability of these markets [5,6]. Several methodological approaches are used to study the influence of speculation on commodity prices. Granger causality tests and price volatility models (such as the GARCH, stochastic volatility modeling, and others) are used to see if speculation measured by trade volume, commercial-to-noncommercial ratios, or other indicators causes commodity prices or volatility. Researchers are also investigating if speculative comovement across markets is related to product and asset links [7,8]. Speculation in the energy market could lead to price spikes and other problems in the grain market because it takes a lot of fuel to make grain.

Typically, less liquid markets, such as livestock products or cotton, have a larger and more statistically significant impact from short-term speculation on return volatility [9]. In addition, Bohl et al. [10] observed a short-term speculation impact on return volatility on rapeseed oil, cotton, sugar, and corn traded on Chinese markets. It can be argued that the inclusion of more speculative indicators and more frequent data can better explain prices. Speculation and its price-distorting effects may be especially common in products heavily impacted by global energy prices and utilized as biofuel. This is particularly true when analyzing markets outside of the United States. For example, according to research conducted by Bandyopadhyay et al. [11], excessive futures market speculation in Indian commodity exchanges increases spot market volatility. Another thing that the results show is that too many short-term investors in the futures market could have a destabilizing impact on these markets.

However, there is less research on European commodity markets that are smaller in size and less liquid or transparent compared to US markets. European agriculture commodity markets, such as the Paris exchange MATIF (Paris, France) and the London exchange LIFFE (London, UK), trade mainly in rapeseed, corn, and milling wheat. Prices in these markets are heavily influenced by commodity prices in the main US markets, but they also attract speculative activity, which may distort pricing during economic turmoil. In their research on products traded on the Paris exchange MATIF, Statnik and Verstraete [12] argue that exogenous factors influence the behavior of agricultural product prices, as reference markets, market depth, and market regulation may all have an impact on market behavior, pointing out short-term memory effects in return volatility. Other, older studies, such as one conducted by Busse et al. [13], argue that the increased European rapeseed price is influenced by speculation, characterized by market over-reactions and high volatilities, and increased correlation with crude oil. On the other hand, more recent studies focus more on structural changes in these markets when trading activity has grown dramatically. Price-shock amplification (period-to-period shock transmission) increased in the Paris and London wheat futures markets after 2006 as trade volume increased [6]. Authors argue that noncommercial positions have been found to stabilize the market during stressful periods. When investigating the London wheat market, Dawson [14] points out a structural change in these markets as the increase in volatility since June 2007 appears not to be short-lived. Futures prices significantly determine volatility, and volatility is stable and highly persistent. Other studies on the European grain futures market focus on relationships between commodities and other financial markets. For example, Makkonen et al. [15] observed that the stock market interacts more with the rapeseed futures market during extreme conditions; moreover, when the economy recovers and the rapeseed market is strong, investors’ positive expectations raise the returns even further. According to Zuppiroli and Revoredo–Giha [16], the US wheat market outperforms European wheat markets in terms of short-term hedging against price movements, making smaller markets more vulnerable to speculative activity and other distortions. This is particularly significant considering the current pandemic-outbreak-caused economic shock.

In the scholarly literature, the influence of the pandemic on agricultural markets has been extensively studied. It has an impact on economic performance, sustainability, and development processes in general [17]. More specifically, health crises such as these have a detrimental impact on the global economy, globalization, food and job security, supply chains, or even food fraud [18]. Stricter government rules and lockdowns, for example, raise concerns about food security as a health and economic well-being problem [19]. Changes in consumer buying behavior, transportation network disruptions, workforce absenteeism, and the closure of major food production businesses have all posed challenges to the food supply chain. [18]. Authors Falkendal et al. [20] point out that production losses have only a modest influence on worldwide pricing and supplies; but trade restrictions and precautionary purchases by a few important players might result in global food price increases and catastrophic local food shortages. Consumer purchase behavior shifted as well and was influenced by income impacts, the opportunity cost of time, and longer planning horizons during the COVID-19 pandemic [18]. According to Coyne [21], negative externalities are caused by infectious illnesses. Market pricing will not represent the social cost of individual activity if these externalities exist, and as a result, market imbalances are probable. Most recent studies on commodity markets during the pandemic period highlight increases in cross-correlations between commodities and increases in hedging and speculative pressures [22,23].

The current COVID-19 situation has had a significant detrimental impact on the European economy in general. In the first quarter of 2020, almost all EU nations saw a drop in exports compared to the previous year [24]. Furthermore, since Western Europe’s agricultural sector is primarily reliant on Eastern European seasonal laborers who work for low rates, the epidemic is driving companies to consider whether this is a sustainable model and if they should instead seek local people [25]. Negative impacts were seen across the board in the agricultural commodities trade, although industries and sectors were affected differently according to their size and kind of product [18]. For example, some farmers who produce particular items (such as grapes and flowers) destroy their unsold supply due to market access issues [26]. As a result, one of the long-term consequences of any crisis is predicted to be a reduction in farmers’ income. COVID-19 also has an influence on how farmers behave. As a result of the drop in agricultural revenue, farmers lowered their crop-related costs [26]. Greater opportunities to hedge against price risks in financial markets may have resulted in better options for dealing with falling prices and income instability.

The COVID-19 pandemic was a shock to present agricultural production and distribution systems, food security, and unemployment rates because of company limitations, and it also resulted in economic instability because of business restrictions [26]. Demand, production, and overall economic activity must be increased to avoid economic stagnation. Therefore, fiscal and monetary policies implemented stimulus packages and announced emergency assistance that were unprecedented in scope and volume, both at the national and European levels [25,27]. A drop in wheat production, together with export restrictions in Russian and Ukrainian wheat markets, is especially important to European agricultural markets. The world’s wheat market is controlled by oligopolistic relationships, with eight nations accounting for 95.6 percent of global exports [28]. Grain prices, on the other hand, remained stable in 2020 due to relatively low energy costs. Researchers on European commodity markets such as Ahmed and Adjemian [29] claim that following 2015, wheat market leadership shifted from the United States to Europe, implying that the French (MATIF) futures market is the primary source of price discovery and therefore leads other markets [29]. Farmers, traders, and other market participants have begun to base their decisions and budgeting on European futures markets rather than US futures markets as a result of changes in the trade map, resulting in the United States losing market leadership in wheat to the former Soviet Union and EU countries [30]. Additional research also indicates that the global wheat market price discovery leadership has shifted from the United States to the French MATIF futures market [31].

To summarize, European markets are utilized in research on occasion, but they may be explored further by adding extra factors to better understand speculation and its influence on agricultural commodity prices or returns. To begin with, these studies lack concrete measures of speculation, such as short-term or long-term speculation indices and their influence on commodity returns. Second, unlike in energy or metal markets, the models provided do not account for seasonality, which is typical in agriculture markets. Finally, greater focus should be placed on the post-2020 era (the COVID-19 pandemic period) when comovement among different commodity types has risen and prices in major commodities markets have become more volatile. The COVID-19 pandemic, which is still ongoing, has had an unprecedentedly huge impact on the lives, societies, economies, and markets of the affected countries [32]. Therefore, the primary goal of this study is to strengthen the other authors’ research into the impact of speculation on agricultural prices and return volatility. Using theoretical and empirical derivatives speculation theories, we study the influence of derivatives speculation on European commodity prices. We also emphasize that in the pandemic period, short-term speculation makes these prices even more volatile.

2. Materials and Methods

2.1. Methods

Besides descriptive statistics, we also employ the Granger noncausality test to study causal linkages between price and returns and speculation, as well as the Augmented Dickey–Fueller test for time series stationarity and the generalized autoregressive conditional heteroskedasticity approach (GARCH) to model price/return volatility. The dependent variable is the continuous futures price or, more specifically, the returns from these futures. We choose, as an independent variable, the short-term speculation index, which is derived as a ratio of total trade volume to open interest.

We begin by defining the variables used in our models. We use the natural log of futures prices to generate a price return series. In futures markets, price indices such as the lowest, highest, opening, and closing prices can be evaluated. We have decided to go with the closing price of the day. Returns on agricultural product futures are calculated using the logarithmic difference between the prices of futures contracts in periods t and t − 1 (Formula (1)). This volatility metric is a logarithmic difference between the futures prices on the futures exchanges used in other authors’ studies, as typically full-time price data is not stationary as compared to futures returns [33,34,35]. The dependent variable returns on the futures contract, represented by $R_t$, will be used for our research. To show returns as a percentage of change, we multiply them by 100:

$$R_t=\ln \left(\frac{P_t}{P_{t-1}}\right)\ast 100.$$

(1)

where: $R_t$ is the return on agricultural futures, $P_t$ is the futures price, $t$ is the time, and $\ln$ is the natural logarithm.

TV/OI is a speculation index that measures short-term speculation (Formula (2)). The main advantage of this speculative index is that it is based on real-time data and is applicable to futures exchanges that do not disclose information on exposure structures, such as European commodity markets. Trade volume shows the intensity of speculative activity, whilst open positions reflect the total amount of hedging activity in commodity markets [10]. As Shear [35] claims, because speculators have a short trading horizon and trade daily, the volume of speculation influences the volume of daily trade. Because this ratio is simple and easy to use, it is often used to show speculative behavior in futures markets.

$$S_t=\frac{T{V}_t}{O{I}_t}.$$

(2)

where: S_t is the short-term speculation index, TV_t is the futures contract trade volume, OI_t is futures contract open interest, and $t$ is the time.

Before we go over the details of our suggested model, it is necessary to assess the stationarity of time series. We then run a test to see if the time series for price, returns, and short-term speculation index TV/OI are stationary. A unit root test is used, both with and without a time trend. When evaluating time series, it is critical that their statistical features and distribution remain constant—especially regarding autocorrelation, mean, and variance. Stationary processes are those that have a constant mean and variation. When conducting causality tests, nonstatistical trends, which are commonly defined by time series in financial markets, may mislead statistical conclusions. To determine whether future variable time series are stationary, the Augmented Dickey–Fuller (ADF) test will be utilized. This approach includes a single-root test to see if time series have a single root that describes nonstationary processes and whether time series have a stochastic trend [36]. An autoregressive time series model (Formula (3)) is the foundation of the ADF test [37]. In this situation, the parameter φ should be equal to 0 for it to be described as having a unit root:

$$\Delta Y_t=\phi Y_{t-1}+u_t.$$

(3)

where: Y_t is the dependent variable return on a futures contracts, φ is a parameter of the model, u_t is the residual error, Δ is the change in the first order, and t is the time.

The supplemented Augmented Dickey–Fueller (ADF) test (Formula (4)) likewise employs a constant, a trend, and a greater number of time lags [38]. This makes it possible to assess whether the time series is stationary by considering (by adjusting the data accordingly) the long-term determinative trend. In this regard, it is assumed that prices will continue to rise over time; therefore, long-term economic growth and pricing changes are removed as a result:

$$\Delta Y_t=\alpha +\beta t+\phi Y_{t-1}+{{\displaystyle \sum }}_{i=1}^j{\theta }_i\Delta Y_{t-i}+u_t.$$

(4)

where: Y_t is the dependent variable return on futures contracts; α, β, φ, and θ are model parameters; u_t is the residual error; Δ is the change in the first order; i is the time lag; j is the number of time lags; and t is the time.

A third-degree root from the sample size will be used to select the number of time lags j = ∛n. Hypotheses for the ADF test can be described as H0: φ = 0, time series have a unit root; H1: φ < 0, time series do not have a unit root.

If the time series in absolute terms of prices does not fit the condition of stationarity, the Granger Causation Test is used to analyze the causal relationship between the specified speculative index and the returns on futures contracts for selected agricultural products. Despite the substantial correlation between the variables, this test will allow the direction of causality to be discovered and assessed to determine whether short-term speculation leads price/returns or vice versa. The Granger Causation Test is expressed as two autoregressive equations (Formulas (5) and (6)). The model’s first equation allows one to check if speculation is not driving prices or returns on a product’s futures contracts (Formula (5)). The model’s second equation allows one to determine whether prices or futures returns do not cause speculation (Formula (6)). Then, it is assessed which time series can better explain the other one under a given number of time lags using the methodology presented by Granger [39]. However, there are times when statistically significant effects with respect to both time series are not detected, or when statistically significant impacts are identified for both time series and the variables are characterized by feedback relationships.

$${\mathrm{Y}}_t={\alpha }_0+{{\displaystyle \sum }}_{i=1}^j{\alpha }_{1i}{Y}_{t-i}+{{\displaystyle \sum }}_{i=1}^j{\alpha }_{2i}{X}_{t-i}+{\epsilon }_t.$$

(5)

$${\mathrm{X}}_t={\beta }_0+{{\displaystyle \sum }}_{i=1}^j{\beta }_{1i}{X}_{t-i}+{{\displaystyle \sum }}_{i=1}^j{\beta }_{2i}{Y}_{t-i}+{\omega }_t.$$

(6)

where: Y_t is the dependent variable return on futures contracts, X_t is an independent variable index of short-term speculative activities, α_0,1,2 and β_0,1,2 are model parameters, ε_t, ω_t are residual errors, i is the time lag, j is the number of time lags, t is the time.

Next, we define our research hypotheses for the causality test:

−

H0: ${\alpha }_{21}={\alpha }_{22}=0$. Speculation does not cause the return on futures contracts.

−

H0: ${\partial }_{21}={\partial }_{22}=0$. The return on futures contracts does not cause speculation.

Then we use GARCH modeling to see how short-term speculation affects price or return conditional volatility. To obtain consistent parameter estimations in GARCH modeling, stationary time series are also necessary [40]. As a result, if the price is not stationary, we employ returns, which are first-order logarithmical price differences. The model consists of mean and variation equations (Formulas (7) and (8)). An autoregressive equation of return from futures contracts is included in the mean equation (Formula (8)). The second equation in the model is called the equation of variation (Formula (8)). It lets us see how the autoregressive link between price/return variability and external (external) variables, such as short-term speculation described by the TV/OI index, affects price/return variability.

This methodology was first described by Engle [41], who proposed ARCH models, and Bollerslev [42], who developed the generalized GARCH methodology. This approach is well suited for financial markets where return volatility is typically clustered and can be split into periods of high or low volatility. The residual error from the mean equation represents innovation and its impact on price and return volatility. Additional exogenous variables that explain agricultural commodity returns can be added into the mean equation. The model’s variance equation also allows us to evaluate the impact of historical variables on the estimated conditional volatility. The residual effect shows if volatility can be explained by its lagging values. Unlike ordinary ARCH, the generalized model GARCH also uses a generalized volatility effect that incorporates multiple lagged residual values from earlier periods. Therefore, it is more user-friendly because it requires a smaller number of parameters to be calculated and taken into consideration. This makes them easier to read and makes the results more explanatory from an economic perspective. In most studies, one autoregressive conditional volatility lag and one generalized conditional volatility lag effect on conditional volatility are used, and therefore such models are named as GARCH (1,1). In our study, we use one-day residuals and volatility lag.

As an extension to our previously published GARCH model, we also use a Threshold Autoregressive Conditional Heteroskedasticity (TGARCH) technique developed by Zakoian [43] for our main model. Using this technique, a binary variable is added to the variance equation to measure the negative return impact on volatility and if the relationship between returns and volatility is asymmetric [44]. This shows that if negative news has a destabilizing effect on markets, it will increase return volatility. Then, we define our preliminary model with one-period AR lags TGARCH (1,1):

Mean equation:

$$R_t={\alpha }_0+{\alpha }_1{R}_{t-1}+u_t.$$

(7)

Variance equation:

$$h_t^2={\beta }_0+{\beta }_1{u}_{t-1}^2+{\beta }_2{h}_{t-1}^2+{\beta }_3{u}_{t-1}^2{d}_{t-1}+{\beta }_4{S}_{t-1}.$$

(8)

where: the mean equation consists of returns $R_t$ as an autoregressive process with parameters ${\alpha }_0$ and ${\alpha }_1$, and an error term $u_t$ with a variance of $h^2$. The conditional variance $h_t^2$ is provided in the variance equation, where ${\beta }_0$ is the constant, ${\beta }_1{u}_{t-1}^2$ is the residual (ARCH) effect, ${\beta }_2{h}_{t-1}^2$ is the variance (GARCH) effect,${\beta }_3{u}_{t-1}^2{d}_{t-1}$ is the asymmetric component, and the parameter $d_t=1$ if $u_{t-1}<0$ and $d_t=0$ otherwise. If ${\beta }_3\ne 0$, then a threshold effect exists; when ${\beta }_3>0$, the return impact on volatility is asymmetrical. We also use an external variable ${\beta }_4{S}_{t-1}$ in the variance equation to assess the direct effect of the speculation on conditional volatility.

Even though spring can be assumed to be the most volatile season for agricultural futures markets, we apply an additional GARCH model to investigate these relationships in more detail for commodities traded in European commodity markets. We then select the exact month for each commodity when their returns are most volatile. To examine seasonal volatility, time seasons described by dummy variables are also included in the GARCH model variance computation. We only utilize 11 months from January to November, to avoid multicollinearity, assuming that agricultural prices are less volatile during winter. The following formulae describe our suggested month-selection model:

Mean equation:

$$R_t={\alpha }_0+{\alpha }_1{R}_{t-1}+u_t.$$

(9)

Variance equation:

$$h_t^2={\beta }_0+{\beta }_1{u}_{t-1}^2+{\beta }_2{h}_{t-1}^2+{{\displaystyle \sum }}_{i=1}^n{\gamma }_i{D}_{it.}$$

(10)

where: the mean equation consists of returns $R_t$ as an autoregressive process with parameters ${\alpha }_0$ and ${\alpha }_1$, and an error term $u_t$ with a variance of $h^2$. The conditional variance $h_t^2$ is provided in the variance equation, where ${\beta }_0$ is the constant, ${\beta }_1{u}_{t-1}^2$ is the residual (ARCH) effect, and ${\beta }_2{h}_{t-1}^2$ is the variance (GARCH) effect. We also use an external variable ${\gamma }_i{D}_{it}$ in the variance equation to assess the direct effect of a month on conditional volatility, where i is the month, and n is the number of months.

Then, using GARCH modeling, we examine how speculative variables influence price or return conditional volatility during the most volatile months. We focus if short-term speculation amplifies this month-related increase in return volatility. For this purpose, we describe how to model short-term speculation as a multiple component with a seasonal dummy variable using a TGARCH model with an extra variable (Formulas (11) and (12)). The impact of seasonally weighted speculation on return volatility can be determined using a similar model but with two additional variables in the variance equation: season effect and season effect multiplied by short-term speculation. This enables us to evaluate the influence and direction of short-term speculation on volatility during the month with the highest volatility after we select the most volatile month for each commodity using our month-selection model (Formulas (9) and (10)). Aside from short-term speculation and seasonality, we employ a one-day lag for autoregressive residual and volatility effects, as well as the dummy variable for asymmetry between returns and conditional volatility, as in the prior example. For clarity, the models are referred to as “Framework I” and “Framework II.” Framework I only examines short-term speculation, but Framework II considers both short-term speculation and seasonal effects. Both Framework I and II use GARCH and TGARCH variants, so there are a total of four models:

Mean equation:

$$R_t={\alpha }_0+{\alpha }_1{R}_{t-1}+u_t.$$

(11)

Variance equation:

$$h_t^2={\beta }_0+{\beta }_1{u}_{t-1}^2+{\beta }_2{h}_{t-1}^2+{\beta }_3{u}_{t-1}^2{d}_{t-1}+{\beta }_4{S}_{t-1}+{\beta }_5{D}_{t-1}+{\beta }_6{D}_{t-1}{S}_{t-1}.$$

(12)

where: ${\beta }_6{D}_{t-1}{S}_{t-1}$ is the combined effect of the speculation index ${\beta }_4{S}_{t-1}$ and the season’s effect ${\beta }_5{D}_{t-1}$. Other parameters are described in the preliminary and main models (Formulas (7)–(10)).

Then, we check if the GARCH/TGARCH model parameters’ p-values are less than 0.05, suggesting statistically significant volatility clustering, effects of exogenous variables, and so on. The following are hypotheses about the effect of speculation on return volatility:

−

H0: ${\beta }_4$ = 0. Speculation has no impact on the volatility of the return.

−

H0: ${\beta }_6$ = 0. Speculation does not amplify return volatility during the most volatile month.

The final step is to compare these models with one another. Therefore, the following information criteria are used to select the best model: the Hannan–Quinn information criterion, the Akaike information criterion, and the Schwartz information criterion (Bayesian). To choose the best model, the information criteria values must be the lowest.

2.2. Data

Using three EU agricultural commodities futures contracts, we study the relationship between changes in commodity prices and speculative activity. More specifically, we investigate how futures speculation measured by a short-term speculative index affects returns and return volatility in milling wheat, corn, and rapeseed futures markets. These futures contracts are traded on the Paris Euronext exchange MATIF. Bloomberg and Barchart provide us with daily closing prices, total open interest, and trading volume through their data platforms [45,46,47]. To measure long-term dynamics in European commodity markets, we use continuous closing prices, which are prices of nearby future contracts, which change on the first trading day of each contract month. Trade volume is the total number of contracts traded during each trading day, whereas open interest shows the total amount of hedging activity in the underlying market. For all three variables, we collected daily data from 23 April 2003, to 1 September 2021. This period saw large price changes as well as considerable increases in open interest throughout our analysis. Commodity prices have risen steadily throughout time, becoming more volatile in recent years (see Figure A1). The rise in commodity prices and volatility has resulted in a significant increase in speculators’ market share in the commodity futures market. Therefore, the rise in commodity prices and volatility has been ascribed to an increase in speculators’ market share. The sample is then separated into two subsamples: the full sample, and after 2020. In the post-2020 period, changes in price patterns and epidemic-induced shocks can be observed. Using these data, we calculate the short-term speculation ratio for each commodity market using trade volume and open interest.

3. Results

We begin with descriptive statistics for rapeseed, corn, and milling wheat futures traded on the Euronext exchange in Paris (MATIF) (see

Table 1. Descriptive statistics of agricultural commodity futures.

Variable	Milling Wheat		Corn		Rapeseed
	Full Sample	Post-2020	Full Sample	Post-2020	Full Sample	Post-2020
Price
Mean	174.470	203.860	170.060	188.600	353.220	431.590
Minimum	103.750	175.250	105.000	160.000	190.500	335.500
Maximum	286.000	254.250	265.000	241.250	577.250	577.250
St. deviation	39.605	17.449	32.642	21.852	80.992	63.571
St. deviation, % mean	22.700	8.559	19.194	11.586	22.930	14.729
Skewnees	0.095	0.572	0.395	0.303	−0.017	0.765
Kurtosis	−0.508	−0.432	−0.172	−1.408	−0.616	−0.809
Return
Mean	0.016	0.060	0.012	0.055	0.019	0.078
Minimum	−18.697	−8.722	−15.553	−15.553	−13.608	−13.608
Maximum	12.507	3.685	7.925	3.390	5.318	3.671
St. deviation	1.299	1.206	1.181	1.192	1.019	1.317
Skewnees	−1.324	−0.861	−1.325	−5.214	−1.674	−2.913
Kurtosis	23.805	6.902	24.338	67.510	17.939	27.195
Speculation index
Mean	0.095	0.128	0.068	0.066	0.085	0.101
Minimum	0.000	0.013	0.000	0.003	0.000	0.017
Maximum	0.396	0.329	0.492	0.249	0.346	0.285
St. deviation	0.054	0.044	0.045	0.035	0.048	0.040
Skewnees	0.960	0.536	1.751	1.315	1.004	0.870
Kurtosis	1.793	0.760	7.044	2.432	1.447	1.385

Source: author’s calculations based on Euronext Commodities (MATIF) data, 2021.

). First, we analyze the full sample data for 2003–2021. The volatility of returns as measured by standard deviation is highest for milling wheat futures (1.299) and lowest for rapeseed futures (1.019). Milling wheat futures have the highest short-term speculative index values (the mean is 0.095), while corn futures have the lowest (0.068).

Next, if we look at the pandemic years of 2020 and after, we can see that the standard deviation of returns is highest for rapeseed (1.317), and it has changed dramatically if compared to full sample results. The standard deviation of returns for corn has remained nearly constant, while it has decreased in milling wheat markets (to 1.206). Short-term speculation index values increased for milling wheat (to 0.128) and rapeseed (to 0.101). However, short-term speculation decreased in the corn market (to 0.066). Mean values of prices are higher in all three commodities during the pandemic years of 2020–2021.

It is also worth noticing that returns do not follow a well-shaped normal distribution. For example, kurtosis is high (>3) for all three commodities using both samples, indicating that many return values are close to the mean or zero. Return skewness is negative for all three commodities in both samples, implying that there are more positive-but-small returns and fewer-but-larger negative returns. To sum up, milling wheat has the highest variation of returns and is the most volatile and risky commodity with the highest short-speculative activity. However, rapeseed futures have changed dramatically in terms of return volatility during the pandemic period, becoming more volatile than milling wheat and having almost the same amount of short-term speculation.

Following that, we present the results of the Augmented Dickey–Fuller (ADF) test using two models: one with only constant and the other with both constant and trend (see

Table 2. Augmented Dickey–Fuller test results.

Time Period	Milling Wheat		Corn		Rapeseed
	Full Sample	Post-2020	Full Sample	Post-2020	Full Sample	Post-2020
Price of commodity
test with constant	0.1133	0.8312	0.0813	0.7778	0.7341	0.9793
with constant and trend	0.1484	0.3689	0.1886	0.3225	0.6749	0.4079
Return of commodity
test with constant	<0.0001	<0.0001	<0.0001	<0.0001	0.0001	<0.0001
with constant and trend	<0.0001	<0.0001	<0.0001	<0.0001	<0.0001	<0.0001
Speculation index of commodity
test with constant	<0.0001	<0.0001	<0.0001	0.0006	<0.0001	<0.0001
with constant and trend	<0.0001	0.0001	<0.0001	0.0046	<0.0001	0.0003

Source: author’s calculations based on Euronext Commodities (MATIF) data, 2021.

). The p-value of price for all three commodities and both ADF models is more than 0.05, indicating that these time series have a unit-root and are nonstationary. However, returns from futures (specified in Formula (1)), which are the first logarithmical difference in price values, have a p-value for all three commodities smaller than 0.05 using both time samples. With a p-value of less than 0.05, the short-term speculation index is also stationary for all three commodities in both time samples. To sum up, all of the time series except for prices are stationary; thus, returns can be properly used for further Granger noncausality investigation. Another thing to keep in mind when using a time–trend model is that the returns are more stationary and the p-values are lower, suggesting that the returns have a time trend throughout this period.

We then present the results of the Granger noncausality test (

Table 3. Estimates of Granger’s noncausality test using one-day, two-day, and combined lags.

Variable/Hypotheses	Full Sample			Post-2020
	Coefficient	p-Value	Result	Coefficient	p-Value	Result
Milling Wheat
H0: the return does not cause speculation:
$\mathrm{The}\mathrm{return}\mathrm{lag}(-1)\mathrm{is}\mathrm{equal}\mathrm{to}\mathrm{zero},{\alpha }_{21}=0$.	0.0007	0.1144	Accept	0.0007	0.1152	Accept
$\mathrm{The}\mathrm{return}\mathrm{lag}(-2)\mathrm{is}\mathrm{equal}\mathrm{to}\mathrm{zero},{\alpha }_{22}=0$.	0.0005	0.2805	Accept	0.0005	0.2852	Accept
$\mathrm{Return}\mathrm{lags}(-1)\mathrm{and}(-2)\mathrm{are}\mathrm{equal}\mathrm{to}\mathrm{zero},{\alpha }_{21}={\alpha }_{22}=0$	-	0.1447	Accept	-	0.1473	Accept
H0: speculation does not cause the return:
$\mathrm{The}\mathrm{speculation}\mathrm{index}\mathrm{lag}(-1)\mathrm{is}\mathrm{equal}\mathrm{to}\mathrm{zero},{\beta }_{21}=0$.	0.4628	0.2986	Accept	0.4607	0.3015	Accept
$\mathrm{The}\mathrm{speculation}\mathrm{index}\mathrm{lag}(-2)\mathrm{is}\mathrm{equal}\mathrm{to}\mathrm{zero},{\beta }_{22}=0$.	0.1891	0.6706	Accept	0.1865	0.6752	Accept
$\mathrm{The}\mathrm{speculation}\mathrm{index}\mathrm{lags}(-1)\mathrm{and}(-2)\mathrm{equal}\mathrm{to}\mathrm{zero},{\beta }_{21}={\beta }_{22}=0$	-	0.2369	Accept	-	0.2427	Accept
Corn
H0: the return does not cause speculation:
$\mathrm{The}\mathrm{return}\mathrm{lag}(-1)\mathrm{is}\mathrm{equal}\mathrm{to}\mathrm{zero},{\alpha }_{21}=0$.	0.0003	0.5462	Accept	0.0020	0.0852	Accept
$\mathrm{The}\mathrm{return}\mathrm{lag}(-2)\mathrm{is}\mathrm{equal}\mathrm{to}\mathrm{zero},{\alpha }_{22}=0$.	0.0003	0.6065	Accept	−0.0003	0.7629	Accept
$\mathrm{Return}\mathrm{lags}(-1)\mathrm{and}(-2)\mathrm{are}\mathrm{equal}\mathrm{to}\mathrm{zero},{\alpha }_{21}={\alpha }_{22}=0$.	-	0.7136	Accept	-	0.2068	Accept
H0: speculation does not cause the return:
$\mathrm{The}\mathrm{speculation}\mathrm{index}\mathrm{lag}(-1)\mathrm{is}\mathrm{equal}\mathrm{to}\mathrm{zero},{\beta }_{21}=0$.	0.6232	0.1328	Accept	2.3764	0.2374	Accept
$\mathrm{The}\mathrm{speculation}\mathrm{index}\mathrm{lag}(-2)\mathrm{is}\mathrm{equal}\mathrm{to}\mathrm{zero},{\beta }_{22}=0$.	0.8441	0.0418	Reject	−1.3586	0.4977	Accept
$\mathrm{The}\mathrm{speculation}\mathrm{index}\mathrm{lags}(-1)\mathrm{and}(-2)\mathrm{equal}\mathrm{to}\mathrm{zero},{\beta }_{21}={\beta }_{22}=0$.	-	0.0053	Reject	-	0.4925	Accept
Rapeseed
H0: the return does not cause speculation:
$\mathrm{The}\mathrm{return}\mathrm{lag}(-1)\mathrm{is}\mathrm{equal}\mathrm{to}\mathrm{zero},{\alpha }_{21}=0$.	−0.0011	0.0405	Reject	−0.0008	0.5374	Accept
$\mathrm{The}\mathrm{return}\mathrm{lag}(-2)\mathrm{is}\mathrm{equal}\mathrm{to}\mathrm{zero},{\alpha }_{22}=0$.	−0.0001	0.9377	Accept	0.0013	0.3117	Accept
$\mathrm{Return}\mathrm{lags}(-1)\mathrm{and}(-2)\mathrm{are}\mathrm{equal}\mathrm{to}\mathrm{zero},{\alpha }_{21}={\alpha }_{22}=0$.	-	0.1178	Accept	-	0.5182	Accept
H0: speculation does not cause the return:
$\mathrm{The}\mathrm{speculation}\mathrm{index}\mathrm{lag}(-1)\mathrm{is}\mathrm{equal}\mathrm{to}\mathrm{zero},{\beta }_{21}=0$.	−0.5440	0.1758	Accept	−1.4491	0.4281	Accept
$\mathrm{The}\mathrm{speculation}\mathrm{index}\mathrm{lag}(-2)\mathrm{is}\mathrm{equal}\mathrm{to}\mathrm{zero},{\beta }_{22}=0$.	0.2808	0.4843	Accept	0.2249	0.9019	Accept
$\mathrm{The}\mathrm{speculation}\mathrm{index}\mathrm{lags}(-1)\mathrm{and}(-2)\mathrm{equal}\mathrm{to}\mathrm{zero},{\beta }_{21}={\beta }_{22}=0$.	-	0.3886	Accept	-	0.6939	Accept

Source: author’s calculations based on Euronext Commodities (MATIF) data, 2021.

). In most cases, the p-value of the underlying AR model is greater than 0.05 for all three commodities. Corn and rapeseed futures are the only two exceptions.

Using a one-day lag, we can reject the hypothesis that returns do not cause speculation in rapeseed futures (p-value is 0.0405). This shows that returns better explain speculation than vice versa. Using a two-day lag, we can reject the hypothesis that speculation does not cause the return in the instance of corn futures (p-value is 0.0418), but we cannot reject the opposite hypothesis (p-value is 0.6065). If we look at the total lag one-directional effect, it is only significant in the corn market (p-value is 0.0053), where returns are better explained by speculation than vice-versa. This shows some evidence of speculation having an impact on returns in corn markets. It is also worth noting that in this scenario, the coefficient values are positive, indicating that speculation increases returns. In the case of the milling wheat market, none of the p-values are above 0.05. This demonstrates that time series are only loosely related to one another. However, the second hypothesis, that speculation does not cause returns, has higher p-values. Other observations are that the p-value is smaller when using a one-day lag, except for corn futures, so more time lags can be added for further investigation.

When using pandemic period p-values for all commodities and time lags, they are all above 0.05. This shows that there is no statistically significant direction from speculation to returns or vice versa. However, p-values during the pandemic period are higher for speculation than returns, showing that returns explain speculation better than vice versa. Even though returns better explain speculation, in the case of corn futures, the opposite is true.

Following that, we investigate the GARCH month-selection model using time dummy variables for months (

Table 4. Estimates from the GARCH month selection model for agricultural commodities.

Month	Milling Wheat	Corn	Rapeseed
$\mathrm{January},{\gamma }_1$	0.0312 *	0.0398 *	−0.0138
$\mathrm{February},{\gamma }_2$	0.0063	0.0333	0.0419
$\mathrm{March},{\gamma }_3$	0.1157	0.0108	0.0649
$\mathrm{April},{\gamma }_4$	0.2950	0.1919	−0.0432
$\mathrm{May},{\gamma }_5$	−0.0187	0.1609 *	−0.0117
$\mathrm{June},{\gamma }_6$	0.0257 *	0.0699	0.0035
$\mathrm{July},{\gamma }_7$	0.0135	0.0857 *	−0.0079
$\mathrm{August},{\gamma }_8$	−0.0051 **	0.0073	−0.0137 **
$\mathrm{September},{\gamma }_9$	0.0042	0.0190	0.0040
$\mathrm{October},{\gamma }_{10}$	0.0131	0.1747 **	−0.0125 *
$\mathrm{November},{\gamma }_{11}$	0.0025	−0.0094	−0.0054

Notes: Estimates with a p-value of less than 0.1 are flagged with one asterisk (*), and those with a p-value of less than 0.05 are flagged with two asterisks (**).

). We look at when these markets are the most volatile. We also take note of cases when p-values are above 0.05 but below 0.10. We are concerned about the p-values for models and their coefficients.

Milling wheat and corn both have a statistically significant effect in January, with a p-value between 0.05 and 0.10 (parameter ${\gamma }_1$ estimations are 0.0312 and 0.0398). However, this effect is relatively small compared to other months. Milling wheat also has a statistically significant impact in June (estimated value is 0.0257) and in August (estimated value is −0.0051, with a p-value below 0.05). However, during April, this effect is estimated to be 0.2950 even though its p-value is higher than 0.10.

On the other hand, corn has a statistically significant and stronger (compared to milling wheat) month’s impact in May (estimated value is 0.1609), July (estimated value is 0.0857), and October (estimated value is 0.1747, with a p-value below 0.05). However, the strongest effect in this market is during April, estimated to be 0.1919, with a p-value above 0.10.

In the rapeseed market, only August (coefficient estimated to be −0.0137) and October (coefficient estimated to be −0.0125) are statistically significant, with p-values below 0.10. However, these values are negative, showing that returns from these futures contracts are less volatile during these months. Rapeseed markets are found to be most volatile in March (coefficient estimated to be 0.0649), but this effect is smaller than it is in milling wheat and corn markets during April and with a p-value above 0.10.

It is evident that all three agricultural commodities have increased return volatility during sowing and before harvest, mostly in the spring months: March–April for milling wheat, April–July for corn, and February–March for rapeseed futures. For further analysis, we select April for milling wheat and corn, and March for rapeseed futures. Even though the p-values are above 0.10 for these months, we will revisit the p-values in our revisited Framework II.

Then, we examine the outcomes of the basic GARCH and threshold TGARCH models to see if speculation has an impact on return conditional volatility as described in our methodology (

Table 5. Estimates of GARCH agricultural commodity-return models using Framework I without a month’s effect.

Time Period	Milling Wheat				Corn				Rapeseed
	GARCH		TGARCH		GARCH		TGARCH		GARCH		TGARCH
	Full Sample	Post-2020	Full Sample	Post-2020	Full Sample	Post-2020	Full Sample	Post-2020	Full Sample	Post-2020	Full Sample	Post-2020
Mean equation:
Constant	−0.0123	0.0154	0.0127 **	0.0584 **	0.0268	0.0310	0.0355 **	0.0906 *	0.0230	0.0696 *	0.0148	0.0381 **
Return	0.0273	0.0407	0.0369 **	0.0797 **	0.1047 **	0.0509	0.0830 **	0.0474	0.1160 **	0.0665	0.0909 **	0.0665 **
Variance equation:
Constant	−0.0083	0.1536	0.0084 *	0.1717	−0.0213 **	−0.0858 *	0.0096	−0.0166	0.0516	0.1035	0.0319	0.0966
Residual	0.2881 **	0.2273	0.0701 **	0.2270 **	0.2072 **	0.1825	0.1678 **	0.0915 **	0.1513 *	0.3547	0.1149	0.2967 **
Volatility	0.7672 **	0.7041 **	0.9493 **	0.7002 **	0.7399 **	0.7600 **	0.8317 **	0.8552 **	0.8040 **	0.6866 **	0.8859 **	0.7553 **
Asymmetry	-	-	−0.1271	−0.3229 *	-	-	0.1966	−0.7400	-	-	0.0839	0.1982
Speculation	0.6645 *	−0.1562	−0.0393	0.0836	2.2663 **	3.4420 **	0.9461 **	1.8786 **	0.1295	−0.5183	−0.0132	−0.4463
Information criteria:
L-lik	−7362	−661	−7265	−658	−6888	−618	−6797	−602	−6344	−645	−6285	−644
BIC	14,775	1358	14,590	1359	13,826	1272	13,653	1247	12,739	1327	12,628	1330
AIC	14,736	1334	14,544	1330	13,787	1248	13,608	1218	12,701	1302	12,583	1301
HQC	14,750	1344	14,560	1341	13,801	1258	13,624	1229	12,714	1312	12,599	1312

Source: author’s calculations based on Euronext Commodities (MATIF) data, 2021. Notes: Estimates with a p-value of less than 0.1 are flagged with one asterisk (*), and those with a p-value of less than 0.05 are flagged with two asterisks (**).

). The TGARCH model shares the same characteristics as the GARCH model, except that it also includes an asymmetry factor (a dummy variable $d_{t-1}$). In this table, we only show GARCH and TGARCH models that are based on Framework I, with only short-term speculation as an exogenous element.

When analyzing full sample data, mean equation parameter values for all three commodities are statistically insignificant or close to zero, which can reflect the fact that these time series are stationary and previous returns do not explain the further ones. Next, we can look further into the variance equation where we put the speculation index as an exogenous factor. Residual volatility is statistically significant (p-value is below 0.05), so we can reject the hypothesis that this parameter is equal to zero. This is present in all cases except for rapeseed when using the TGARCH model (estimated value of lagged residual volatility to current volatility is 0.1149). This shows that volatility closely reflects its lagged values, as evidenced by residuals. The volatility effect is statistically significant for all three commodities, indicating that their return volatility is clustered. In other words, the market activity timeline can be grouped into high and low volatility periods. For all three commodities, the asymmetry coefficient is nonsignificant, indicating that there is no asymmetry for positive or negative return to increase volatility. Constants are close to zero or statistically insignificant in both mean and variance equations. When using the GARCH model in the milling wheat market and both the GARCH and TGARCH models in the corn market, the speculation effect on volatility is statistically significant and increases volatility. This effect is higher in the corn market than in milling wheat; it is especially high when using the basic GARCH model (estimated value is 2.2663). In the milling wheat market, this effect is only significant under a p-value greater than 0.05 and lower than 0.10. When modeling rapeseed returns, the information criteria are discovered to be the lowest. When modeling the milling wheat market, the information criteria are found to be the highest. The information criteria are smaller for TGARCH estimates.

When using the post-2020 data, we see that only TGARCH model estimates show statistically significant mean equation estimations (p-values are below 0.05) in the mean equation for both milling wheat and rapeseed futures. The residual is statistically significant only when using the TGARCH model and for all three commodities. Volatility is statistically significant for all three commodities using both models. The asymmetry factor, unlike in full sample data, is statistically significant in the milling wheat futures market but is negative (−0.3229), indicating that positive returns are followed by increased volatility. The speculation index is only statistically significant in the corn market using both models (estimations are 3.4420 and 1.8786). As we see, this effect is stronger than compared to full sample results. The information criteria for the corn market are the smallest, while those for milling wheat are the largest. Information criteria for GARCH and TGARCH models are similar, but are smaller in TGARCH models. To sum up, speculation increases volatility in the corn market using both models, and this effect has become stronger during the post-2020 era.

Following that, we provide further estimates with extra exogenous variables put into the mean equation, which may better explain movements of agricultural commodity returns (see

Table A1. Estimates of GARCH agricultural commodity-return models with additional variables using Framework I without a month’s effect.

Time Period	Milling Wheat				Corn				Rapeseed
	GARCH		TGARCH		GARCH		TGARCH		GARCH		TGARCH
	Full Sample	Post-2020	Full Sample	Post-2020	Full Sample	Post-2020	Full Sample	Post-2020	Full Sample	Post-2020	Full Sample	Post-2020
Mean equation:
Constant	−0.0169	0.0536	0.0181	0.0999 *	−0.0049	0.0768	0.0260 **	0.1159 **	0.0167	0.0540	0.0150 **	0.0274
Return	0.0315	0.0205	0.0446	0.0644	0.1141 **	0.0760	0.0957 **	0.0332	0.1129 **	0.0785	0.0892 **	0.1016 **
GSCI Energy	0.0288 **	0.0204	0.0240 **	0.0135	0.0235 **	0.0180	0.0284 **	0.0189	0.0503 **	0.0338 **	0.0539 **	0.0355 **
S&P 500	0.0221	−0.0152	−0.0153	−0.0043	−0.0152	0.0357	−0.0152	0.0510	0.0033	0.0291	−0.012 **	0.0248
Euro Stoxx 50	0.0144	0.0351	0.0558 **	0.0315	0.0656 **	0.0550	0.1028 **	0.0239	0.1085 **	0.1007 **	0.1148 **	0.1078 **
3M Eurodollar	−0.0384	−5.968 **	−0.1255	−5.102 **	0.2310	−3.390 **	0.8789 **	−2.5027	−0.3647	−4.378 **	−0.420 **	−4.181 **
Variance equation:
Constant	−0.0079	0.0611	0.0068	0.1840 **	−0.0025	0.0658	0.0052	−0.0049	0.0501	0.0715	0.0115	0.0530
Residual	0.2866 **	0.0627	0.0653 **	0.2156 **	0.2078 **	0.9183	0.1882 **	0.1081 **	0.1548 **	0.3633	0.0790	0.2878 **
Volatility	0.7643 **	0.9104 **	0.9529 **	0.7227 **	0.7732 **	0.2345	0.8202 **	0.8538 **	0.8050 **	0.6988 **	0.9324 **	0.7761 **
Asymmetry	-	-	−0.1673	−0.433 **	-	-	0.1591 **	−0.659 **	-	-	−0.0210	0.1934
Speculation	0.7092	−0.1677	−0.0295	−0.1725	1.0773 **	3.4771	0.9572 **	1.4974 **	0.0852	−0.4008	−0.0164	−0.2087
Information criteria:
L-lik	−7344	−640	−7247	−650	−6134	−600	−6725	−592	−6215	−612	−6144	−611
BIC	14,774	1346	14,586	1367	12,373	1260	13,543	1250	12,515	1285	12,380	1289
AIC	14,709	1302	14,515	1323	12,302	1219	13,472	1206	12,451	1244	12,309	1244
HQC	14,732	1319	14,540	1340	12,327	1235	13,497	1223	12,473	1260	12,334	1262

Source: author’s calculations based on Euronext Commodities (MATIF) data, 2021. Notes: Estimates with a p-value of less than 0.1 are flagged with one asterisk (*), and those with a p-value of less than 0.05 are flagged with two asterisks (**). S&P500, GSCI Energy, Euro Stoxx 50, 3-month Eurodollar indices are used as Δlog by subtracting the index value logarithms.

). We employ log differences of the S&P 500, GSCI Energy, Euro Stoxx 50, and 3-month Eurodollar indexes to depict the economic environment, namely, economic growth and energy prices. This leads to some important observations. The GSCI energy index exhibits a statistically significant influence with p-values of less than 0.05 for all three products when considering the whole sample and for rapeseed futures when analyzing post-2020 data as well. Coefficient values are positive, indicating that rising energy prices enhance agricultural futures returns and vice versa. This demonstrates that energy costs have a significant impact on agricultural commodity returns. In more cases than the S&P 500 index, the Euro Stoxx 50 is statistically significant and has positive coefficients. In all circumstances, the Euro Stoxx 50 index is statistically significant in the rapeseed market. This demonstrates that the Eurozone stock market has a greater influence on agricultural prices than the S&P 500, which is composed of companies based in the United States. This suggests that rising European stock market returns are correlated with rising agricultural commodity prices traded on the MATIF. The 3-month Eurodollar index is statistically significant and has negative coefficients except for the corn market when analyzing full sample data. The higher the implied 3-month U.S. dollar LIBOR interest rate, the lower this index value. Therefore, when interest rates grow, returns from agricultural commodity futures grow as well, and vice versa. Other estimates are comparable to those in models where there are no extra variables in the mean equation. Even though all models indicate statistically significant volatility effects, a statistically significant asymmetry component is detected in the corn market (the coefficient is 0.1591 when analyzing the full sample and −0.659 when analyzing post-2020 data). However, in the corn market, short-run speculation is merely statistically significant. When both techniques and time samples are used, the information criterion for all three commodities is slightly smaller.

Then, we examine the outcomes of the GARCH and TGARCH models to see if speculation has an impact on price conditional volatility when the month is also taken into consideration (

Table 6. Estimates of GARCH commodity-return models using Framework II with a month’s effect.

Time Period	Milling Wheat		Corn		Rapeseed
	GARCH	TGARCH	GARCH	TGARCH	GARCH	TGARCH
Mean equation:
Constant	−0.0071	0.0016	−0.0016	0.0329 **	0.0231	0.0208 **
Return	0.0621 **	0.0580 **	0.1121 **	0.0919 **	0.1077 **	0.0943 **
Variance equation:
Constant	−0.0029	0.0150	−0.0024	0.0111	0.0309	0.0223 *
Residual	0.1546 **	0.1113 **	0.1972 **	0.1398 **	0.0753 **	0.0795 **
Volatility	0.8541 **	0.9094 **	0.7878 **	0.8547 **	0.9035 **	0.9236 **
Asymmetry coefficient	-	−0.0963	-	0.2031 **	-	0.0278
Speculation index (1)	0.2508	−0.0212	0.9835 **	0.6988 **	−0.1326	−0.0899 *
Month (2)	0.3982	0.1495	−0.0351	−0.1679 **	−0.1364 **	−0.0575 *
1 × 2	−0.6940	−0.4433	0.4559	4.1949 **	2.8548 **	1.1661 **
L-lik	−7234	−7139	−6180	−6720	−6262	−6200
BIC	14,535	14,354	12,436	13,515	12,591	12,475
AIC	14,483	14,295	12,378	13,457	12,540	12,417
HQC	14,501	14,316	12,398	13,478	12,558	12,438

Source: author’s calculations based on Euronext Commodities (MATIF) data, 2021. Notes: Estimates with a p-value of less than 0.1 are flagged with one asterisk (*), and those with a p-value of less than 0.05 are flagged with two asterisks (**).

). These models have an additional two exogenous variables: a dummy variable $D_{t-1}$ representing the most volatile month, and speculation multiplied by this dummy variable $D_{t-1}\times S_{t-1}$. As the month-selection model shows, milling wheat and corn futures returns are the most volatile in April, and rapeseed futures returns are the most volatile in March.

As in our previous models based on Framework I, mean equation parameters are close to zero, yet here they are more statistically significant. The residual and volatility effects from the variance equation are statistically significant in all cases, including rapeseed. This again shows that return volatility is clustered in these markets and that current volatility closely follows its previous values. Then again, constants are close to zero or statistically insignificant in both mean and variance equations. The asymmetry factor is only significant in the corn market (estimated to be 0.2031), showing that negative returns are followed by increased volatility. Negative news affects corn futures volatility when using this improved model. Next, we analyze the impact of exogenous factors on return volatility. Neither speculation nor month had a statistically significant effect on milling wheat returns, even though milling wheat, when analyzing descriptive statistics, was found to be the most volatile and had the highest speculation-index mean value. Speculation increases return volatility throughout the year in the corn market using both the GARCH model (parameter estimation is 0.9835) and the TGARCH model (parameter estimation is 0.6988). Month and combined effect are statistically significant only in TGARCH estimates, and season speculation amplifies return volatility (parameter estimation is 4.1949) while month alone reduces volatility (parameter estimation is −0.1679). In the rapeseed market, speculation reduces volatility throughout the years, but this effect is only statistically significant in the TGARCH model (which is estimated to be negative −0.0899). Month reduces volatility (effects are estimated to be −0.1364 and −0.0575) and the combined effect increases volatility (effects are estimated to be 2.8548 and 1.1661) in both models.

Information criteria are estimated to be the smallest when modelling corn returns with the GARCH approach and rapeseed returns with the TGARCH approach, and largest when modelling the milling wheat market. GARCH estimates have a lower information value only in the corn market. The information criteria are smaller than in the previous model for all three commodities. We focus on these results because the GARCH model for corn has a lower information criterion. In the rapeseed market, there is evidence that during more volatile time periods, speculation increases volatility. Seasonal volatility is amplified in the corn market as well as the rapeseed market.

Following that, we present estimates for Framework II models that integrate economic variables such as the S&P 500, GSCI Energy, Euro Stoxx 50, and 3-month Eurodollar indices into the mean equation (see

Table A2. Estimates of GARCH commodity-return models with additional variables using Framework II with a month’s effect.

Time Period	Milling Wheat		Corn		Rapeseed
	GARCH	TGARCH	GARCH	TGARCH	GARCH	TGARCH
Mean equation:
Constant	−0.0115	−0.0038	−0.0048	0.0245 **	0.0186	0.0229 **
Return	0.0615 **	0.0621 **	0.1143 **	0.0966	0.1044 **	0.0948 **
Δlog GSCI-Energy	0.0214 *	0.0191 **	0.0233 **	0.0297 **	0.0515 **	0.0546 **
Δlog S&P 500	0.0159	−0.0002	−0.0150	−0.0043 **	−0.0001	−0.0057
Δlog Euro Stoxx 50	0.0178	0.0373 **	0.0652 **	0.0777 **	0.1091 **	0.1110 **
Δlog 3 m Eurodollar	−0.0593	−0.0473	0.2271	0.8577 **	−0.4183	−0.3880 **
Variance equation:
Constant	−0.0018	0.0137 **	−0.0017	0.0090 **	0.0296 *	0.0162 **
Residual	0.1575 **	0.1078 **	0.2080 **	0.1527 **	0.0709 **	0.0697 **
Volatility	0.8501 **	0.9125 **	0.7725 **	0.8442 **	0.9086 **	0.9360 **
Asymmetry coefficient	-	−0.1192 **	-	0.1859 **	-	−0.0501
Speculation index (1)	0.2618	−0.0179	1.0834 **	0.7572 **	−0.1387	−0.0783 **
Month (2)	0.3903	0.1424 **	−0.0220	−0.1720 **	−0.1257 **	−0.0467 *
1 × 2	−0.5673	−0.4103 **	0.2579	4.1686 **	2.5993 **	0.9439 **
L-lik	−7223	−7126	−6134	−6668	−6131	−6057
BIC	14,548	14,362	12,389	13,445	12,363	12,225
AIC	14,470	14,278	12,305	13,362	12,286	12,141
HQC	14,498	14,307	12,335	13,391	12,313	12,170

Source: author’s calculations based on Euronext Commodities (MATIF) data, 2021. Notes: Estimates with a p-value of less than 0.1 are flagged with one asterisk (*), and those with a p-value of less than 0.05 are flagged with two asterisks (**).

). The GSCI energy index was statistically significant for all three items except wheat when using the GARCH model, with a p-value of less than 0.10. Positive coefficient values indicate that increasing energy prices improve agriculture futures returns and vice versa. When using the GARCH approach and having positive coefficients, the Euro Stoxx 50 is statistically significant in all cases except the wheat market. When employing the TGARCH technique, the S&P 500 index is only statistically significant in the corn market but has a negative coefficient. Only in the maize and rapeseed markets, particularly when using the TGARCH model, is the 3-month Eurodollar index statistically significant. The coefficient value in the corn market is positive, but it is negative in the rapeseed market. Other estimates are comparable to those in models where there are no extra variables in the mean equation. However, when employing the TGARCH technique in the wheat market, there is a statistically significant influence from short-run speculation multiplied by month on returns (−0.4103). Short-run speculation, in this sense, lessens volatility during the more volatile month of April. Asymmetry may also be seen in the wheat market. However, it has a negative coefficient (−0.1192). Most of the time, when both methodologies and time samples are used, the information criteria values are slightly smaller.

We conclude in the following section that GARCH approaches can be effectively used to analyze realized futures returns in European commodity markets. Time series are stationary and, in most cases, residual and volatility effects are present under a p-value of 0.05. This shows that returns are clustered, and volatility follows its lagged values. Therefore, European agricultural futures trading activity can be split into periods of high and low volatility. The asymmetry factor has no or mixed results, as it is only statistically significant when using a month-based model for the corn market or a basic model for milling wheat during the pandemic period. This shows that negative information is not necessarily destabilizing these markets. Corn markets, on the other hand, showed good evidence that speculation was having a significant and growing effect on return volatility.

4. Discussion

4.1. Contextualization with Previous Research

These GARCH-modeling findings, when combined with the Granger noncausality test, provide a compelling case. Even though similar methods were applied by other authors, they used either the US markets or different dummy variables [48,49,50]. As a result, our study came up with three major results:

First, not all products investigated in this paper are statistically significantly affected by short-term speculation. Many other authors came up with similar results. Typically, neither or only some of the commodities used in research have shown some form of relationship [5,51,52]. Thus, our research results show that the price bubble cannot be blamed directly on speculative factors. This is opposed to, for example, Adämmer and Bohl [53], who proposed that speculative bubbles were frequent in the wheat markets in 2006–2008. Furthermore, increased speculation may have explained some of the price increases in the soybean markets [54]. However, this impact is mostly related to data analysis up to 2010. Therefore, similarly to Etienne et al. [55], we cannot agree that during the present time, agricultural futures markets were characterized by price bubbles that were driven by speculative forces. On the other hand, speculative influences are debated, in much research, to have at least a minimal impact on prices. According to [56,57], it has been demonstrated that having a higher number of speculative variables increases the accuracy of price volatility predictions. In our study, even though corn futures were not the most volatile or had the highest mean value of speculative index, they are the only product where speculation can explain both return and return volatility.

Second, in our research, corn futures returns are found to be partially driven by short-term speculation. However, in much of the research conducted by other authors, opposing or insignificant effects are observed, and agricultural returns are better explained by returns [36,52,58,59]. However, similar effects from speculation on corn returns were observed in some studies, but these studies are typically older. For example, in the years 2006–2008, speculative price bubbles were discovered in the wheat and corn futures markets [53]. According to a study conducted by Shanmugah and Armah [60], except in the corn and cattle markets, index fund holdings cannot be considered to influence prices. However, when evaluating the influence of speculation on price fluctuations, the wheat markets may be regarded as an exception since it cannot be ruled out that speculation drives prices rather than the other way around [3]. However, as noticed in our research, this effect on corn markets is not true when analyzing the post-2020 period. The current research on the pandemic period highlights the importance of energy markets and their relationships with the corn market, which is used as biofuel. Speculation in energy and precious metal futures is more prevalent during crisis periods and even more so during the COVID-19 pandemic. In contrast, agricultural futures attract more hedging pressure [22]. The cross-correlations of multifractality between crude oil and the sugar future market are the strongest, and the cross-correlations of all the agricultural futures increased after the emergence of COVID-19, except for the orange juice future market [23].

And finally, corn return volatility is also driven by short-term speculation, whereas other commodities have mixed or no results. These results are similar to research conducted by Bohl et al. [10], where the return from corn futures in non-US markets was found to be driven by short-term speculation. In addition, research conducted by Bandyopadhyay et al. [11] also provides evidence that excessive speculation in futures markets increases spot market volatility, therefore suggesting that the excess presence of short-term investors can destabilize the futures market. On the other hand, some researchers point out that commodities such as corn may have had a stabilizing effect on return volatility. For example, speculation in soybean futures markets, which are likewise less liquid, has stabilized prices since 2003 [50]. In corn markets, a stabilizing impact of speculation on price volatility has been established, as has been shown in prior research by other authors [5,48]. When it comes to prices, except for corn after 2006, speculation cannot be said to influence them [60]. In our research, only the rapeseed futures market showed some evidence of a stabilizing effect. Our research also provides indications that the effect of speculation on return volatility is amplified in the corn market during the more volatile month of April. In other writers’ research, seasonality has a substantial influence on the volatility of agricultural commodity prices, and the highest volatility is often in the month before harvest, with US maize futures swinging at their peak in May and lowest in November [61,62]. Karali and Power [63] propose that individual product price volatility and structural components are better explained by specific causes than by macroeconomic variables, which they observed during the period of 2006–2009 for different agricultural products. Financial turmoil during periods of crisis may well amplify return volatility. For example, during the 2008–2009 crisis, goods prices, especially metals and energy items, were highly intertwined, and speculation in this area might have resulted in price synchronization across various items [64]. During the 2008–2011 crisis, the total volatility of energy and maize prices reached up to 45 percent [65]. Similar effects can be observed during the pandemic period of 2020–2021.

4.2. Research Limitations and Further Research Guidelines

This leads to the conclusion that corn futures are mostly affected by short-term speculation and that the destabilizing effect can only be argued in the corn market. Therefore, future research should also emphasize its comovement with energy prices if both short-term speculation and return volatility are driven by changes in energy prices. Energy price movements are analyzed, especially in contemporary research. For example, Hung [66] finds a strong comovement between crude oil prices and agricultural commodity markets predominantly during the COVID-19 outbreak compared to the preCOVID-19 period. According to Wen et al. [67] during the outbreak of COVID-19, the spillover effect of the stock market on the commodity market has been significantly enhanced. The effect of biofuel, together with speculation on European corn prices, should be analyzed as well, as there are many fewer studies on this subject in European markets compared to the US. Even though, according to older studies [68], energy market speculation stabilizes prices more than maize speculation, according to Etienne et al. [51], because maize is used as a biofuel to produce ethanol, corn prices have grown increasingly tightly related to energy product prices, and this external influence accounts for more than 30% of the variance. However, according to Cao and Cheng [69], the food-oil market system has the strongest spillover effect in the short term, and the spillovers during the pandemic are significantly weaker than those during the financial crisis. According to Fan et al. [70], the cross-speculative pressure remains relatively low, and the increased speculation does not cause seemingly unrelated commodities to become correlated. Further research shows that when crude oil futures prices go down, speculation helps to lessen the negative effects of positive macroeconomic uncertainty changes on futures returns [71].

Most studies on EU markets are relatively dated [12,13]. Price shock amplification rose in the Paris and London wheat futures markets in 2006 when these markets grew, resulting in market over-reactions and excessive volatility, as well as a stronger link with crude oil [6]. Dawson [14] points out that the rise in volatility since June 2007 looks to reflect a long-term structural shift in these markets. Studies may as well employ other commodities, such as rapeseed oil or meal. For example, the findings of Lawson [72] reveals that the impact of speculation on product prices varies depending on the commodity (rice and wheat prices are less responsive than maize and soybean prices) and the variable used to represent speculation. For example, according to Živkov [73], oil and soybean futures are the best diversification tools since their prices are least reliant on oil prices and their natural volatility and surges. Furthermore, more emphasis should be put on factors that describe long-term speculation and compare them to short-term speculation. Long-term speculation indices are analyzed by many other authors, but not in the EU markets. According to Manera et al. [74], long-term speculation increases volatility, whereas short-term speculation decreases it. Ludwig [75] notes that long-term speculation, which includes a positioning structure, gives liquidity to markets over time, but short-term speculative consumption depletes liquidity, necessitating more study using daily trade volume data. Short-term fluctuations in open interest might be primarily driven by speculators’ demand for liquidity. Therefore, speculators, as identified by the money managers category of the CFTC, may be responsible for increasing volatility in several markets (corn, wheat, soybean oil, coffee, and cotton) [76]. Except for the wheat markets, the volatility of the noncommercial position in corn markets has decreased. In research of the wheat and maize markets, a similar influence on volatility was discovered by Borin and Di Nino [77].

Once more information becomes available on noncommercial positions on the Euronext exchange in Paris (MATIF), excessive speculation measured by the Working T index can be calculated. This index is usually used in studies on US markets [78], and it can be used to improve our proposed model and better explain speculation in European corn markets. A speculation index based on the most popular and extensively used Working T Index by other authors, which demonstrates excessive speculating, was employed in an empirical investigation [79]. In addition, the GARCH model can employ more dummy variables, such as ones explaining shock moments or structural breaks. An important observation is that in the post-2003 period, when financialization processes developed and interactions between financial markets intensified, the effect of seasonality on return volatility was weaker than in previous periods. Various writers looked at negative and positive return effects, as well as asymmetric connections [74,80,81]. Baur and Dimpfl [80] found that agricultural commodity futures markets showed similar asymmetric relationships when using the TGARCH approach: positive asymmetry factors were seen mostly in maize markets, and in wheat markets they were generally negative. As a result, a good return is more likely to increase price volatility than the other way around. Other authors employed the GARCH (DCC) approach and observed that nonfundamental factors, such as commodity market financialization and market sentiment, play important roles in driving return comovement over the sample period, though their impacts vary over time [82]. More sophisticated GARCHs such as EGARCH or APGARCH can be included for long-term memory effects in return volatility similarly to studies by Czudaj [49]. On the other hand, the GARCH-M model for predicting risk premiums might be used to further characterize these relationships [83]. Continuous Granger causality tests may be employed as well [84]. The fact that the residual errors of the models are correlated shows that these interactions are nonlinear, allowing the Granger causal test findings to be used in more advanced approaches, such as that conducted by Dick and Panchenko [85]. Finally, more observations during the pandemic and postpandemic periods can be added to the calculations once more time passes.

Another important observation is that GARCH modeling occasionally produces contradictory findings between the complete sample and the post-2020 data. This might be related to GARCH and TGARCH’s failure to detect abrupt changes in regimes. As a result, in future research, a more dedicated tool, such as Regime-Switching or more sophisticated structural break models, can be used to see how short-run speculation affects agricultural commodity returns in response to changing regimes reflecting major shifts in the economic environment, not just in the pandemic period. Improved models may also provide a more exact definition of the duration of these modes and when exactly these structural changes happen.

4.3. The Practical Significance of the Study

Many researchers investigate the impacts of trade restrictions on commodity pricing and price volatility and how to quantify these effects [49,86,87]. Producers and dealers who are more sensitive to price risk and are willing to face uncontrollable financial losses are the ones that choose product futures [88]. Another important consideration is that commodity traders who are more vulnerable to price shocks may require more stringent restrictions on speculative activity in commodity markets [89]. Even those scholars who have documented specific instances in which speculative factors have destabilized pricing believe that bans or other limitations would be harmful to market growth at this stage of product market creation [35]. For example, Acharya et al. [9] developed a model that permitted the evaluation that noncommercial market players’ capital limits impose restraints on commercial market participants’ ability to manage price risk, affecting futures and product prices. Others argue that hedging pressures, price volatility, liquidity, and the risk premium will be more distorted by position limits [90]. On the other hand, there is a great significance of providing market players and policymakers with accurate information on the different categories of traders and their relevance to the market’s and pricing mechanism’s effective operation [75].

Therefore, active and passive measures should be distinguished. Active mechanisms include position limitations in US and EU product markets, trading day limits for market price fluctuations, extra transaction fees for trading transactions, marginal account requirements, and other marginal requirements. Passive measures include tougher reporting requirements, more product market transparency, and tighter regulation of over-the-counter trade. Active measures, according to empirical research, would not accomplish the stated purpose and would further destabilize pricing. Therefore, the use of passive measures, on the other hand, is more reasonable considering our work and the findings of other empirical investigations [5,87].

Commodity exchanges should completely segment market participants into commercial and noncommercial players, as well as give higher frequency data, more thorough data on trade processes, and market concentration indicators. Increased product financial market openness and a clearer legal architecture would go a long way toward lowering product market uncertainty and pricing volatility [78,91]. This is especially true for commodity exchanges in other countries, where, owing to a lack of data and the inability to detect all markers of speculative behavior, only a small number of academics have employed these marketplaces in their empirical study. Market participants and regulators should be informed of the results.

A well-functioning commodities market with liquid futures contracts may help in resolving the various problems caused by pandemics and other exogenous shocks. A well-performing commodity futures market, for example, provides an opportunity to hedge against price risks that are common in agriculture, especially during epidemic periods marked by considerable price volatility and farmers’ income uncertainty. Second, commodity futures markets benefit not only farmers but also agricultural producers who hold long positions to hedge against rising prices, thereby preserving supply chains of agricultural goods, which, as others have noted, are vulnerable to health crises such as these and contribute to food security. Finally, futures contracts may be used during times of uncertainty, such as rising energy prices, so their use is still beneficial when the epidemic period is over.

5. Conclusions

This research investigates the above-mentioned connection for the Paris Exchange MATIF, which is motivated by disagreement among empirical results in the literature concerning the stabilizing or destabilizing influence of speculative activity in futures markets. We used realized, daily returns on rapeseed, milling wheat, and corn futures traded on the European commodity exchange MATIF. We investigated data from 2003 to 2021. This time encompasses various events connected to financialization and commodities market globalization, as well as more than a year of the pandemic period. We observed that these commodities have increased in their return volatility, or speculative activity, over this time. The speculation index, which is calculated by dividing trade volume by open interest, is a proxy for speculative behavior in our research. We, like many other authors, use this speculation measure based on the assumption that speculators engage in short-term trading activity attempting to gain profits from price changes. In our research, we analyze the volatility of three agricultural commodities using extended autoregressive conditional heteroskedasticity GARCH models as well as Granger noncausality testing. Seasonal effects, and whether speculation makes returns more volatile during volatile months, were added to the GARCH model. Dummy variables were also added to the model.

Our study provides three important findings. First, we uncover evidence that short-term speculation drives corn market returns; moreover, speculation causes these markets to be more volatile. Corn markets, on the other hand, are neither more volatile nor have higher levels of short-term speculation than milling wheat or rapeseed. Second, the influence of short-term speculation on return volatility in the corn market has risen over the pandemic era, indicating that speculation may have skewed this market during the COVID-19-induced economic shock. Finally, there are insights that this influence is exacerbated in the corn market during the more volatile month of April since this month is known to be the most volatile, and more new information enters these markets considering that season’s crop. However, according to our study, there is not enough data to back up the destabilizing hypothesis for all agricultural commodities.

Our study’s results have important policy implications. Because of financial speculation, futures commodity exchange regulators have proposed limiting trading activities. Our results, like those of other authors, indicate that financial speculation has a limited influence on price levels and volatility in agricultural markets and that, in certain cases, speculators help to bring new information and correct prices. Another thing to take into consideration is that restrictions on commodity trading can make these markets less liquid and prevent them from effectively hedging against price risks. However, we demonstrate that if short-term speculation is destabilizing these markets, this effect is only observable in corn markets. Therefore, it should be investigated whether energy costs impact not just corn prices, but also encourage speculation, and whether the connectivity of the corn and oil markets makes them more vulnerable to adverse speculative repercussions. Future studies should focus more on long-term speculation and its effects on the return volatility on European agricultural commodity markets once more information about noncommercial traders’ positions becomes available.