Does Sentiments Impact the Returns of Commodity Derivatives? An Evidence from Multi-commodity Exchange India

Abstract

The advancements in technology, increased accessibility to various modes and platforms of communication, and increased willingness on the part of participants to share their ideas/opinions has resulted in huge amounts of data on the World Wide Web, hence, easily available to impact decision-making. Furthermore, commodity prices are primarily driven by demand and supply, wherein such news is open to the cognitive thinking of individuals. Thus, using the principles of natural language processing, which combines concepts of linguistics, computer science and artificial intelligence, helps in improving the accuracy of price determination. Therefore, this article aims to examine the relationship between sentiments conveyed through various sources and the performance of India’s largest commodity market, multi-commodity exchange (MCX). The correlation and causation between sentiment scores extracted from such textual content and the daily returns of select commodity derivatives are analysed. The results show varying levels of significance of sentiments on the returns of commodity contracts and imply that there is an increased scope of using such unstructured content in the field of finance.

Keywords

Behavioural Finance Sentiment Analysis Lexicon-based Approach Natural Language Processing Futures Commodity Contracts

Introduction

Financial decisions made by individuals were seen to be purely rational where utility maximization, given income or profit maximization, was the key consideration. However, over time, with the occurrence of several unprecedented economic events that could not be fully explained using the existing knowledge, it became vital to look into various other aspects that could influence the decision-making of individuals. Here, behavioural finance and related developments helped in better explaining the deviations from rational behaviour. For instance, taking into account cognitive biases resulting in asymmetric information processing, which when applied in economic decision-making is termed as prospect theory in behavioural finance and is widely studied (Alfano et al., 2015). Studies, in this regard, conclude that decision-making is prone to bounded rationality as individuals are subject to several behavioural biases, or there are limitations in terms of their cognitive ability such as being susceptible to emotions, prior knowledge, and limited mental and computational capacity (Altman, 2010; Baker & Nofsinger, 2011; Duxbury, 2015; Uzar & Akkaya, 2013). Also, studies indicate an increased exposure to sentiments, wherein one’s own opinions or views can be subdued by certain mental states that are specific to the decision-maker, resulting in the adoption of simpler choice rules, which align with one’s reasonable heuristics, thus increasing risks like ‘sentiment risk’. Sentiment risk is defined as ‘the effect of the psychological state of mind wherein emotions and current state of mood result in irrational trading or overreaction or under reaction in markets’ (Soon, 2010). Therefore, it eventually deteriorates the pricing quality and increases the chances of incurring losses (Walliser, 2007). Given this, studies show that combining concepts from diverse disciplines such as behavioural finance, cognitive economics, computer science, linguistics, etc., help in better understanding the financial markets by uncovering multiple factors that drive financial concepts like return and risk, eventually enabling improved accuracy of price forecasting and facilitating better decision-making (Tseng et al., 2018).

Further, the recent multifarious developments in media and communications have enabled the faster dissemination of information, especially in financial markets (Fang & Peress, 2009; Peress, 2014), which are highly volatile, uncertain, complex and ambiguous. Also, a drastic shift is seen in the adoption of media to newer forms, such as social media, blogs or online news sites, and in the willingness among the users to express their opinions on these platforms. These changes have led to wider reach and faster inclusion of huge amounts of information and related sentiments in the decision-making process. Subsequently, sentiments are increasingly analysed using sentiment analysis, which comes within the ambit of natural language processing and is seen as an important factor influencing decision-making process—essentially a deliberation process wherein various input factors or information from the environment are broken down into smaller steps (Walliser, 2007). Also, the Global Sentiment Analysis Software Market report states that the global use of sentiment analysis is to grow at a compound annual growth rate (CAGR) of 16.20% during the period from 2017 to 2021 (Research and Markets, 2017). This growth is attributed to the increasing use of sentiment analysis in customer relations and in the engagement of stakeholders of several industries to achieve digital transformation. Digital transformation is progressively adopted in various sectors, ranging from retail to telecom, banking, financial services and insurance (BFSI) to education and has enabled in improved branding and increased profits.

Research Gap Addressed

Given the backdrop, this article aims to understand whether sentiments conveyed by content on digital news platforms and social networking sites have an impact on the performance of commodity derivatives traded on the largest Indian commodity market, namely the multi-commodity exchange (MCX). The relevance of this research comes as researchers (Bloomfield, 2010; Uzar & Akkaya, 2013) have been able to conclude that human cognition is influenced by emotions, intuitions or prior experiences and, thus, impact decision-making. Financial decisions are no different, and, hence, financial markets can be better understood using behavioural aspects, which have not been captured in models for a long time, since they are not objective and not easily measurable. However, in recent times, several developments like sentiment analysis, among others, combine the concepts of computer science, linguistics and several other fields, therefore, enabling researchers and practitioners across the world in detecting and capturing patterns, sentiments and emotions conveyed through several forms of content (Fan et al., 2014; Chen et al., 2014; Li et al., 2014; Tetlock, 2007). Also, this study essentially focuses on the commodity markets since the effect of the news on this market is seen to be more profound than that observed for the stock market (Borovkova, 2015) and, thus, influencing the trading strategies of individuals and non-traditional financial players such as hedge, investment and pension funds over the recent times.

This article is further structured with a review of relevant literature regarding the use of qualitative data for financial decision-making. It is followed by a description of the methodology adopted for the conduct of this research and the findings arrived at. The conclusion derived from this study helps to culminate the article with insights regarding the implications of this study and how this study can be further enhanced.

Literature Review

Wider and relatively easier access to multiple sources of information, shared platforms and tools developed for data collection and its storage, coupled with creative innovations such as content analysis, data mining and several others, have enabled researchers and industry practitioners to detect patterns and sentiments conveyed through various content generated by individuals and institutions. Furthermore, gauging the sentiments conveyed through various sources is important as it is a vital component influencing cognitive thinking and financial decisions and is, hence, increasingly analysed within the field of behavioural finance. Many researchers (Dash & Mahakud, 2012, 2013; Gotthelf & Uhl, 2019; Tetlock et al., 2008), through their works, have concluded it is vital to include sentiment risk in explaining asset returns, as the traditional risk factors cannot be considered as a comprehensive list of determinants of financial market behaviour, by analysing multiple pricing models.

Several empirical studies conducted by Aggarwal (2017), Gao et al. (2020), Rupande et al. (2019), Ryu et al., (2020) and Yelamanchili (2019) have proved the relationship between stock market returns and investor sentiments captured through proxies and by comparing the frequency of a company in news and media to its stock returns (Alanyali et al., 2013). Increasingly, sentiments are being extracted by focusing on the linguistics of the content instead of being captured by proxies alone. This has been possible due to developments like sentiment analysis, thus facilitating the incorporation of sentiments and emotion in financial analysis too. The works of Deshmukh et al. (2016), Khedr et al. (2017), Li et al. (2009), McGurk et al. (2020) and Mo et al. (2016), prove that asset price has a close correlation with information sentiment, captured using sentiment analysis, and that a large volume of news leads to a noticeable increase in the volatility trend. Furthermore, the possibility of combining patterns and the sentiments extracted from content has also been explored by researchers such as Rao and Srivastava (2013) and Zhou et al. (2018), etc. The results showed that different emotions have varying causal relations with asset prices and have proved the possibility of using sentiments as a variable for understanding and forecasting stock market returns, eventually aiding in devising profitable trading strategies.

Alternatively, examining the impact of sentiments on the commodity markets is also important, since commodity prices are primarily driven by news about supply and demand, such as Organization of the Petroleum Exporting Countries (OPEC) or inventory announcements, geopolitical, weather-related and other external news. In this regard, Alfano et al. (2015) test the relevance of prospect theory, a behavioural finance concept, which states that negative information outweighs positive information, in influencing oil prices. The methodology adopted involves sentiment analysis of textual data obtained from Thomson Reuters News Archive to investigate the differential effect of sentiments at various return levels for different types of investors, namely the informed and the uninformed investors. The key results derived from the analysis are that investors process news sentiments asymmetrically with news sentiments having a stronger effect on noise traders. Also, it was found that noise traders have stronger reactions to negative news sentiments than to positive news. Researchers (Li et al., 2016) were also able to conclude that there is a significant correlation between the sentiment series and the price series, and that sentiment of the financial information helps determine the directional movements of the oil prices, since it was observed that the sentiment series changes direction, in terms of peaks and valleys, before the price series. Further, causal relationship was found between sentiments and prices, thus indicating the predictive capacity of sentiments on asset prices. Therefore, using sentiments for forecasting, results in higher prediction accuracy of models. Shen et al. (2017) is another important study drawing empirical evidence regarding how investors’ emotions affect the commodity market returns and whether emotions can be used as a variable for predicting future prices of gold, crude oil and Commodity Research Bureau (CRB) Index commodity market index. The findings from the paper suggest that emotions such as fear, optimism and joy have significant influence and can help predict individual commodity returns but not the market index. Furthermore, empirical results support and show the presence of sentimental effect and appraisal effect in the spot markets of crude oil and gold.

Latest studies by authors (Chen et al., 2020; Jin et al., 2020; McGurk et al., 2020; Su & Li, 2020; Xia et al., 2020; Yadav & Vishwakarma, 2020) show as to how sentiment analysis can be effectively coupled with complex models such as the deep learning architectures, long short-term memory (LTSM)-based model for prediction, attention mechanism and empirical mode decomposition (EMD) method, which is generally used for decomposition of signals. The works have explored several facets, such as the effectiveness of sentiment analysis in classifying the sentiments conveyed, using different languages, and as to whether a reduction of sentences to emotional expression improves the accuracy of the model coupled with further analysing and comparing the direction and intensity of return spillover, volatility and sentiment spillover across different financial markets. Thus, the authors reiterated the fact that timely and accurate prediction of asset prices is vital to investor choice and is important for the national economic stability, as more reasonable regulation can be formulated to guide financial markets, ranging from stocks to cryptocurrencies, which prove to be a cornerstone for the sustainable development of any economy.

Therefore, understanding how sentiments have an impact on the financial markets is seen as a common factor in all papers, wherein different tools and techniques have been used to gauge sentiments and to analyse its relationship with the performance of different markets and assets. However, to the best of the knowledge of the authors, not many papers have considered a large number of commodity derivative contracts to obtain a comprehensive view of any commodity market and the impact of sentiment scores on the performance of the commodity derivatives. Therefore, this article aims at advancing knowledge in data-driven modelling, factoring everyday news and content easily accessed by a majority of traders and investors.

Methodology

To determine the relationship between sentiments and performance of the commodity exchange, the quantified scoring of the sentiments conveyed through various sources and the returns of commodities are considered as variables for the study. An overview of the data and its sources, the selected commodity exchange, the variables used and the period for which the data are collected, along with the research methods and empirical tests to be applied on the data sets are discussed next.

Data and Basis for Selection

To conduct this study, commodity returns and sentiments conveyed through various sources have been analysed. The population for this study is all commodity derivatives traded at the MCX. A sample of 7 out of the 18 generic commodities has been selected based on the highest volume and value of contracts traded on MCX (Money Control, n.d.). Being a non-random sampling technique, the minimum sample size required is not set but is that size that ensures the needed information is obtained. Thus, to ensure a comprehensive and concise study of the exchange, gold and silver have been selected from bullion, nickel and copper from metals and cotton and crude palm oil (CPO) from agro-commodities, while crude oil is selected from the energy segment of the market.

Description of the Commodity Exchange

For this study, the MCX has been selected to examine the causal relationship of sentiments and returns. This selection is based on the fact that MCX is India’s first listed exchange to offer commodity option contracts across various segments of commodities. The MCX boasts of an extensive national reach with its presence in around 1,010 cities and towns across India as of 31 March 2020. Furthermore, with a market share of 94.01% in terms of the value of commodity futures contracts traded (Multi Commodity Exchange India, n.d.) and a turnover of ₹84 trillion in the financial year 2020 (The Hindu Business Line, 2020), MCX is India’s leading commodity derivatives exchange. Thus, due to its wider reach and various avenues available for trading, it is probable that a large number of traders and investors are interested in MCX.

Sources of Data

Textual data such as news, opinions of experts and tweets have been captured to provide a complete set of information without bias and as available to individual or small-scale traders. The commodity prices are collected from the MCX website, while news articles and expert opinions are collected based on relevant keywords from money control and economic times based on the online following for these sources, while opinions of peers are obtained from Twitter.

Period of the Study

The study is for a period of 3 years and 3 months, from 1 January 2017 to 31 March 2020. Content from news websites and Twitter and daily closing prices of the latest commodity futures (FUTCOM) contracts of the selected commodities are collected for the given period.

As presented in Table 1, the latest three contracts each are considered for gold, silver and crude oil based on their expiry, while, for the other four commodities, five to seven contracts each have been used for analysis. For instance, for CPO, two contracts each have been considered for a year for analysis since they are 6-month contracts. But few contracts are 5-month contracts, while few are 7-month contracts; hence, contracts are chosen keeping in mind the tenure to ensure data are collected for the entire period.

Table 1.

Commodities and Contracts Used for the Study

S. No.	Commodity	FUTCOM Contract	Period Covered
1	Gold	February	16/2/2017 to 5/2/2020
		December	2/1/2017 to 5/12/2019
		October	2/1/2017 to 4/10/2019
2	Silver	March	16/3/2017 to 5/3/2020
		December	2/1/2017 to 5/12/2019
		September	2/1/2017 to 5/9/2019
3	Crude oil	February	22/8/2017 to 17/2/2020
		January	20/7/2017 to 17/1/2020
		December	20/6/2017 to 18/12/2019
4	Cotton	February 2020	2/1/2017 to 28/2/2020
		August 2019
		February 2019
		October 2018
		April 2018
		October 2017
		April 2017
5	CPO	February 2020	2/1/2017 to 28/2/2020
		September 2019
		April 2019
		November 2018
		June 2018
		January 2018
		August 2017
		March 2017
6	Copper	January 2020	2/1/2017 to 17/1/2020
		July 2019
		January 2019
		July 2018
		January 2018
		July 2017
		January 2017
7	Nickel	February 2020	2/1/2017 to 28/2/2020
		September 2019
		April 2019
		November 2018
		June 2018
		January 2018
		August 2017
		March 2017

Source: The authors.

Variables for the Study

Sentiment scores are obtained by considering daily news articles, expert opinions and tweets regarding commodities as a single string is taken as the independent variable, while the daily returns of the commodity is taken as the dependent variable.

Returns are calculated from the daily closing prices of commodities, which are collected from the MCX website using Equation (1):

R_{t} = \ln (\frac{P_{1}}{P_{0}}) \times 100

(1)

where, P₁ = today’s closing price of the respective commodity, P₀ = previous day’s closing price of the respective commodity and ln = natural logarithm.

Analytical Tools

Analytical tests such as correlation, regression and Granger causality are carried out using EViews to explain how sentiments bring about a change in the performance of the commodity market, considering returns of commodities and sentiment scores, which are used as is initially and then represented as dummy variables D1 (=1 if the sentiment is positive, else 0) and D2 (=1 if the sentiment is negative, else 0).

Equation (2) shows how sentiment scores impact the performance of the commodities as analysed using regression analysis:

R {(c o m \mod i t y)}_{t} = β_{0} + (β_{1} \times s e n t i m e n t_{t}) + ε_{t}

(2)

where R(commodity)_t = return on day t for a given commodity, sentiment_t = sentiment score conveyed through content from select sources for day t, β₀ = regression constant, β₁ = regression coefficient and ε_t = error term.

Equation (3) shows how sentiment scoring, represented by dummy variables D1 and D2, impacts the performance of the commodities as analysed using regression analysis:

R {(c o m \mod i t y)}_{t} = (β_{1} \times D_{1}) + (β_{2} \times D_{2}) + ε_{t}

(3)

where, R(commodity)t = return on day t for a given commodity, D1 and D2 = dummy variables, β1 and β2 = regression coefficients, ε_t = error term. (The regression constant was dropped to avoid dummy variable trap.)

Equations (4) and (5) show the Granger causation or the cause and effect relationship between sentiment scores of news articles, tweets, expert opinions and the performance of the commodities. Further, the Granger causality test reveals the unidirectional and bidirectional causal relationship between the performance of the commodities and the sentiment scores. The test is applied to the time series data: sentiments scores computed on a day-to-day basis and daily returns.

y_{t} = α + \sum_{i = 1}^{n} β_{i} y_{t - i} + ε_{t}

(4)

y_{t} = α + \sum_{i = 1}^{n} β_{i} y_{t - i} + \sum_{i = 1}^{n} γ_{i} x_{t - i} + ε_{t}

(5)

where y_t = return on day t for a given commodity, x_t = sentiment score conveyed through the sources at time t, α = constant, β, γ = coefficients and ε_t = error term.

Sentiment analysis facilitates in measuring emotional expressions in consistent and quantifiable terms to further understand as to how publicly available data influence multiple happenings or phenomena (Rambocas & Pacheco, 2018). Figure 1 graphically presents the process where sentiments are analysed using a lexicon-based approach and further analysed against returns using correlation and Granger causality as in the work by Tabari et al. (2018), which is detailed further. To analyse correlation and causation, textual data are converted into numerical sentiment scores. Textual content in the form of tweets is captured through Twitter APIs, while news websites are scrolled through, and relevant content like news articles and expert opinions are collected using Beautiful Soup in Python by passing key terms. The raw data scraped from the websites consist of HTML links, tags, texts and timestamp, which is cleaned, to obtain relevant text/string against each timestamp. Further, the timestamp is trimmed to get the date, and the text is combined to arrive at the text for each date instead of time. Post the collection and processing of the input, the unstructured data are pre-processed by splitting each string into tokens or individual words, removal of stop words such as ‘at’, ‘in’, ‘was’, etc., and irrelevant characters like ‘$’, ‘@’, etc., and reducing words to their root form (e.g., falling to fall). Further, using lexicon dictionaries, sentiment scores (from −1 to 1) are obtained, which convey the sentiments of the collected text that is classified into positive, negative and neutral sentiment by assigning 1/−1 and 0 as codes, respectively.

Figure 1.

Source: The authors.

Discussion and Findings

The initial phase of analysis involves the examination of the descriptive statistics of the variables of the study.

Table 2a clearly presents the descriptive statistics of the daily returns of three contracts each of silver, gold and crude oil. The average daily returns for all commodity futures are close to 0 throughout the study period. The silver September contract has the highest mean return of 0.0535, while the silver March contract has the lowest mean return of 0.0159 but with the largest deviation from the mean of 1.0747. The lowest deviation from the mean of 0.7500 is for the gold October contract. Among the crude oil contracts, the February–August contract has the lowest mean of 0.0357, while the December–June contract has the highest mean of 0.0563, with the lowest standard deviation of 1.8740. The highest standard deviation of 1.9469 is for the January–July contract.

Table 2a.

Descriptive Statistics of the Daily Returns of the Latest three FUTCOM Contracts of Gold, Silver and Crude Oil traded on MCX

	Silver March	Silver December	Silver September	Gold February	Gold December	Gold October	Crude Oil February to August	Crude Oil December to June	Crude Oil January to July
Mean	0.0159	0.0251	0.0535	0.0455	0.0476	0.0515	0.0357	0.0563	0.0522
Median	0	0	0	0.0318	0.0098	0.0258	0.0748	0.1677	0.1110
Maximum	6.7490	11.0282	8.2820	4.0791	5.5502	6.6446	12.2660	13.7822	12.7298
Minimum	–4.8835	–3.8437	–4.0742	–2.9024	–2.3692	–2.4355	–7.4338	–8.9994	–12.2921
Std. dev.	1.0747	1.0598	0.9832	0.7529	0.7527	0.7500	1.8740	1.9446	1.9469
Skewness	0.1651	2.0645	1.3935	0.6900	0.8136	1.5160	0.0659	0.1810	–0.2616
Kurtosis	7.6202	23.7387	13.8648	7.2271	8.6079	15.1822	7.5116	9.0593	9.1099
Jacque–Bera	584.66	11104.08	2788.83	546.22	846.73	3637.89	563.6253	951.8693	1004.3600
Probability	0	0	0	0	0	0	0	0	0
Observations	654	596	532	663	596	554	664	620	641

Source: The authors.

Table 2b clearly presents the descriptive statistics of the daily returns of cotton, CPO, nickel and copper. Among the FUTCOM metal contracts, nickel has a higher mean of 0.0609 with the largest deviation from the mean of 1.6700, while copper has a mean of 0.0158 and a standard deviation of 1.0575. Among the agro-commodity contracts, cotton has a mean of −0.0099, while returns showing a deviation of 1.1633 from the mean. CPO has a mean of 0.0453 and standard deviation of 0.9469.

Table 2b.

Descriptive Statistics of the Daily Returns of the latest FUTCOM contracts of Agro-commodities and Metals traded on MCX

	Cotton	Crude Palm Oil	Nickel	Copper
Mean	–0.0099	0.0453	0.0609	0.0158
Median	0	0	0	–0.0114
Maximum	10.4098	4.0333	8.0529	6.7797
Minimum	–11.8430	–3.2968	–7.8722	–4.4279
Std. dev	1.1633	0.9469	1.6700	1.0575
Skewness	–1.2620	0.3646	0.2374	0.6614
Kurtosis	48.4183	4.6896	5.4463	7.6350
Jacque–Bera	54057.6800	94.5370	173.3560	648.5962
Probability	0	0	0	0
Observations	627	670	670	670

Source: The authors.

Taking all the contracts, the descriptive statistics also show that the return distributions are not symmetric since the estimated coefficients for the skewness of the return series are different from 0. The estimated coefficients for the kurtosis of the daily return series are relatively high, implying that the distributions are leptokurtic or heavily tailed when compared to a normal distribution. The observed skewness and kurtosis indicate that the distributions of daily returns are non-normal. The Jarque–Bera normality test also shows the non-normality of the return distributions, with the estimated values of the Jarque–Bera statistic of all the return series being statistically significant at the 1% level.

Table 3a clearly shows the descriptive statistics of the daily sentiments conveyed by several sources while considering silver, gold and crude oil. The gold February contract shows the highest mean sentiment score of 0.0453, with the highest standard deviation of 0.0922. While the silver December contract has the lowest mean sentiment score of 0.0304, its standard deviation from the mean is 0.0410. Among the crude oil contracts, the sentiment scores corresponding to the February–August contract has the lowest mean of 0.0302 with a deviation of 0.0423. The sentiment scores of January–July contract have the highest mean of 0.0313 with a standard deviation of 0.0411.

Table 3a.

Descriptive Statistics of the Daily Sentiments Considered for the Latest three FUTCOM Contracts of Gold and Silver Traded on MCX

	Silver March	Silver December	Silver September	Gold February	Gold December	Gold October	Crude Oil February to August	Crude Oil December to June	Crude Oil January to July
Mean	0.0303	0.0304	0.0315	0.0453	0.0304	0.0306	0.0302	0.0309	0.0313
Median	0.0308	0.0307	0.0312	0.0323	0.0307	0.0308	0.0307	0.0310	0.0311
Maximum	0.1509	0.1509	0.1509	0.7324	0.1509	0.1509	0.1509	0.1509	0.1509
Minimum	–0.2286	–0.2222	–0.2222	–0.2286	–0.2222	–0.2222	–0.2286	–0.2222	–0.2222
Std. dev	0.0427	0.0410	0.0387	0.0922	0.0410	0.0402	0.0423	0.0410	0.0411
Skewness	–0.8905	–0.7241	–0.6689	4.1131	–0.7241	–0.6444	–0.8924	–0.6920	–0.6560
Kurtosis	7.4378	6.5490	7.3237	25.9362	6.5490	6.6673	7.5867	6.4447	6.3130
Jacque–Bera	623.0832	364.8747	454.0703	16402.0800	364.8748	348.7810	670.1877	356.0192	339.1260
Probability	0	0	0	0	0	0	0	0	0
Observations	654	596	532	663	596	554	664	620	641

Source: The authors.

Table 3b clearly presents the descriptive statistics of the daily sentiments while considering cotton, CPO, nickel and copper. Cotton has a mean sentiment score of 0.0294 with a standard deviation of 0.0414, while CPO, nickel and copper have a mean sentiment score of 0.03 and a deviation from the mean of 0.0424 because the contracts have the same execution dates. For all commodities, except cotton, the dates for the contracts have been the same and hence the same descriptive statistics.

Table 3b.

Descriptive Statistics of the Daily Sentiments Considered for the Latest FUTCOM Contracts of Agro-commodities and Metals traded on MCX

	Cotton	CPO	Nickel	Copper
Mean	0.0294	0.0300	0.0300	0.0300
Median	0.0296	0.0305	0.0305	0.0305
Maximum	0.1509	0.1509	0.1509	0.1509
Minimum	–0.2286	–0.2286	–0.2286	–0.2286
Std. dev	0.0414	0.0424	0.0424	0.0424
Skewness	–0.6520	–0.8741	–0.8741	–0.8741
Kurtosis	6.3668	7.4862	7.4862	7.4862
Jacque–Bera	340.5617	647.1640	647.1640	647.1640
Probability	0	0	0	0
Observations	627	670	670	670

Source: The authors.

Table 4 presents the results of unit root test, which is carried out to check for whether the time series commodity returns data are stationary. This is a prerequisite for conducting additional tests to examine the relationship between the variables. It is evident that the daily returns of all the contracts of all the commodities traded on MCX taken for this study are stationary at level since the p-values of the augmented Dickey–Fuller test (ADF) and Phillips–Perron (PP) tests are <0.05.

Table 4.

Unit Root Test results of Daily Returns of Commodities Traded on MCX

Commodity	Method	ADF Fisher Chi-square	PP–Fisher Chi-square
Silver March	Statistic	229.3280*	353.2130*
	Probability	0	0
Silver December	Statistic	239.2660*	350.2330*
	Probability	0	0
Silver September	Statistic	152.0220*	324.2780*
	Probability	0	0
Gold February	Statistic	183.2340*	283.4620*
	Probability	0	0
Gold December	Statistic	220.4470*	331.1740*
	Probability	0	0
Gold October	Statistic	231.5990*	345.6730*
	Probability	0	0
Crude oil December to June	Statistic	114.8000*	318.5860*
	Probability	0	0
Crude oil Feb–Aug	Statistic	151.9150*	343.5060*
	Probability	0	0
Crude oil January to July	Statistic	119.3250*	335.6150*
	Probability	0	0
Cotton	Statistic	227.3520*	351.5800*
	Probability	0	0
CPO	Statistic	237.0160*	355.8600*
	Probability	0	0
Nickel	Statistic	215.5160*	337.5530*
	Probability	0	0
Copper	Statistic	224.1530*	346.0140*
	Probability	0	0

Source: The authors.

Note: * Indicates at 1% level.

Table 5 presents the correlation coefficients, which help determine the strength and nature of the linear relationship between the dependent and independent variables. Table 5 clearly indicates that for all the contracts of the various commodities considered, there is a positive relationship between daily returns and sentiments, except for the gold contract with February 2020 as expiry. The daily returns of this contract and the daily sentiments show a low negative correlation of 0.0040, which implies that a fall (rise) in the sentiment scores leads to an increase (decrease) in daily returns of the contract. Further, the correlation coefficients being closer to 0 show that there is low correlation or weak impact of sentiments on daily returns.

Table 5.

Cross Correlation Between Daily Returns and Sentiment Scores of Commodities Traded on MCX

Variable	Correlation Coefficient
Silver March	0.0681
Silver December	0.0410
Silver September	0.1164
Gold February	–0.0040
Gold December	0.0637
Gold October	0.1127
Crude oil February to August	0.0632
Crude oil December to June	0.0839
Crude oil January to July	0.0900
Cotton	0.0066
CPO	0.0736
Nickel	0.0981
Copper	0.0536

Source: The authors.

Table 6a presents the results of the regression analysis carried out to draw the relationship between returns and sentiment scores taken as is. Due to p-values being less than 0.05, we can conclude that there is a significant impact of sentiments on the daily returns of the September and March FUTCOM contracts of silver and December and October FUTCOM contracts of gold. This indicates that for these contracts alone, the daily sentiments can be considered as a factor, which can help model the returns for the day. While for all other contracts, it is seen that sentiment scores do not have a significant impact on daily returns. This implies that news regarding improved demand for precious metals during the festival seasons can bring about an impact in the returns of commodity futures, especially when these precious metals are seen as safe havens as mentioned in the work by Chevallier et al. (2014). However, the work suggests that gold has a different reaction to economic news when compared to silver, which can be seen in this work too. Meanwhile as concluded by Chevallier et al. (2014), industrial metals, energy and agricultural commodities showcase complex relationship with economic news flow even in this study, with the latter two influenced highly by local factors.

Table 6a.

Result of Regression Analysis Taking Sentiment Scores

Commodity	No. of Observations	Coefficient	t-Statistic	Probability	R²
Silver March	654	3.4034	2.5692	0.0104*	0.0113
Silver December	596	2.3988	1.6293	0.1038	0.0050
Silver September	532	3.3346	2.2497	0.0249*	0.0139
Gold February	663	–0.1246	–0.3607	0.7184	0.0006
Gold December	596	2.3128	2.1959	0.0285*	0.0085
Gold October	554	2.5638	2.3245	0.0205*	0.0134
Crude oil February to August	664	–1.0039	–0.4198	0.6747	0.0102
Crude oil December to June	620	0.6922	0.2610	0.7942	0.0105
Crude oil January to –July	641	1.3273	0.5087	0.6112	0.0096
Cotton	627	1.3698	0.8517	0.3947	0.0024
CPO	670	1.4112	1.1726	0.2414	0.0061
Nickel	670	3.0568	1.4445	0.1491	0.0095
Copper	670	1.0655	0.7962	0.4262	0.0019

Source: The authors.

Note: * Indicates at 5% level.

Table 6b presents the results of regression analysis carried out to draw a relationship between returns and sentiments, which are classified into positive and negative. For this, dummy variables D1 and D2 are considered, where D1 is 1 if the sentiment is positive, else 0 if the sentiment is negative, and D2 is 1 if the sentiment is negative, else 0 if the sentiment is positive. From the results, we can conclude that there is a significant impact of negative sentiments on the daily returns of the silver contract, which expired in March 2020 and crude oil contract with February 2020 and August 2019 as expiry periods. This implies that with every negative sentiment becoming positive, there is an increase in the returns of these contracts, and the results are consistent with the findings of Chevallier et al. (2014), which mention that energy markets have a weak reaction to economic news, while silver reacts positively (negatively) to negative (positive) surprises when we consider most of the business indicators.

Table 6b.

Result of Regression Analysis Taking Dummy Variables to Measure Sentiment Scores

Commodity	No. of Observations	Variable	Coef.	t-Statistic	Probability	R²
Silver March	654	D1	–0.1230	–1.6433	0.1008	0.0113
		D2	0.2297	2.0627	0.0396*
Silver December	596	D1	–0.0801	–0.9915	0.3219	0.0050
		D2	0.1654	1.3492	0.1778
Silver September	532	D1	–0.0601	–0.7589	0.4483	0.0139
		D2	0.0109	0.0866	0.9310
Gold February	663	D1	0.0619	1.5846	0.1135	0.0006
		D2	0.0057	0.0736	0.9413
Gold December	596	D1	–0.0520	–0.8998	0.3686	0.0085
		D2	0.1477	1.6838	0.0928
Gold October	554	D1	–0.0389	–0.6511	0.5153	0.0134
		D2	0.0393	0.4251	0.6709
Crude oil February to August	664	D1	0.1844	1.3793	0.1683	0.0102
		D2	–0.4034	–1.9852	0.0476*
Crude oil December to June	620	D1	0.1335	0.9112	0.3626	0.0105
Crude oil December to June		D2	–3.5976	–1.6158	0.1067
Crude oil January to July	641	D1	0.1032	0.7107	0.4775	0.0096
		D2	–0.3108	–1.4142	0.1578
Cotton	627	D1	0.1343	1.0368	0.3003	0.0024
		D2	–0.0796	–0.9020	0.3674
CPO	670	D1	0.0080	0.1189	0.9054	0.0061
		D2	–0.0416	–0.4025	0.6874
Nickel	670	D1	–0.0038	–0.0321	0.9744	0.0095
		D2	–0.1220	–0.6710	0.5025
Copper	670	D1	–0.0005	–0.0071	0.9943	0.0019
		D2	–0.0043	–0.0374	0.9702

Source: The authors.

Note: * Indicates at 5% level.

Further, Table 7 presents the results of the Granger causality test that is run to determine the cause and effect relationship between the two variables. It is clearly observed that there is a unidirectional causality that daily returns of copper, nickel, CPO and the February and August biannual contracts of crude oil have on the sentiments conveyed through various sources. This shows that the changes in the daily returns of these contracts bring about the changes in the sentiments conveyed through various sources regarding the commodities market. Also, there is no evidence to support the existence of any other unidirectional or any bidirectional causality between daily returns and sentiment scores, implying that sentiments conveyed through the sources considered for this research do not have direct causation on the daily returns of the contracts considered for this sutdy. The findings are consistent with the results of the studies conducted by Bognár (2016), Johnman et al. (2019) and Mo et al. (2016), where the feedback effect between market returns and sentiments derived from news articles is analysed. The results showed that the market movement has an immediate and pronounced impact on the news sentiments, while there is a delayed impact of news on the market movement.

Table 7.

Result of Granger Causality tests between Daily Returns and Daily Sentiment Scores

Commodity	Null Hypothesis	F-statistic	Probability
Silver December	Sentiment scores does not GC returns	1.8248	0.1622
	Returns does not GC sentiment scores	2.5185	0.0815
Silver September	Sentiment scores does not GC returns	2.1286	0.1200
	Returns does not GC sentiment scores	0.3726	0.6891
Silver March	Sentiment scores does not GC returns	2.0325	0.1319
	Returns does not GC sentiment scores	3.0034	0.0503
Gold February	Sentiment scores does not GC returns	1.6922	0.1849
	Returns does not GC sentiment scores	0.4039	0.6679
Gold December	Sentiment scores does not GC returns	1.9793	0.1391
	Returns does not GC sentiment scores	0.2415	0.7855
Gold October	Sentiment scores does not GC returns	2.0013	0.1362
	Returns does not GC sentiment scores	0.4961	0.6092
Crude oil December to June	Sentiment scores does not GC returns	1.9042	0.1498
	Returns does not GC sentiment scores	0.4050	0.6672
Crude oil February to Aug	Sentiment scores does not GC returns	2.8644	0.0577
	Returns does not GC sentiment scores	3.4070*	0.0337
Crude oil January to July	Sentiment scores does not GC returns	1.4068	0.2457
	Returns does not GC sentiment scores	0.7906	0.4540
Cotton	Sentiment scores does not GC returns	0.6083	0.5446
	Returns does not GC sentiment scores	2.1713	0.1149
CPO	Sentiment scores does not GC returns	2.6252	0.0732
	Returns does not GC sentiment scores	9.2525*	0.0001
Nickel	Sentiment scores does not GC returns	2.3199	0.0991
	Returns does not GC sentiment scores	3.8967*	0.0208
Copper	Sentiment scores does not GC returns	0.1552	0.8562
	Returns does not GC sentiment scores	6.0134*	0.0026

Source: The authors.

Note: * Indicates at 5% level.

Conclusion

The study aimed to understand whether there is a significant impact of sentiments conveyed through tweets, newspaper articles and expert opinion on the performance of commodity derivatives traded at MCX and, further, to find any causal relation between sentiments and the performance of commodity derivatives traded on MCX. The results show a significant impact of the sentiments conveyed through the various sources on the September and March futures contracts of silver and December and October futures contracts of gold traded on MCX. However, for other commodities, it is observed that sentiments have taken a back seat while trading, primarily because the extent of coverage of these commodities in the generic public sources considered in this study is limited, as found by closely observing the data set. This is also supported by Ferguson et al. (2015) and Tetlock et al. (2008) in their works, which concluded that data sources read by more sophisticated investors tend to have a significant effect on the markets. Further, it was observed that a large amount of news related to commodities captured for this study does not portray a strong positive or negative outlook to significantly influence returns. Also, at several instances (Johnman et al., 2019; Mo et al., 2016), it was observed that market sentiment is a reaction to the market prices and hence cannot be of help in predicting future stock returns but has an effect on the asset’s volatility caused due to additional noise in the market. This facet of returns influencing market sentiments was seen in many contracts included in this study such as copper, nickel, CPO and the February and August biannual contracts of crude oil. Therefore, keeping a track of such unstructured yet crucial data helps traders to tap on the earnings and safeguard from the losses brought about by market dynamics, shaped by participants’ behaviour, influenced by such widely used sources of information.

Limitations of the Study

Being a developing futures market, the Indian commodity market is largely dependent on the global market conditions and other significant macroeconomic indicators (Mukherjee & Goswami, 2017). Therefore, using more advanced approaches of sentiment analysis and including a larger pool of both general and specific commodity-related information can further enhance this study, which aims to be a foundation for further research in data-driven or algorithm trading in commodity by incorporating behavioural data to increase accuracy. Furthermore, lags of a few days can be introduced in analysing the time series, thus enhancing further research regarding how sentiments influence the performance of derivatives of commodities. Such advancements enable sophisticated institutional traders who increasingly use event-driven arbitrage strategies to make decisions by complementing existing prediction models. Additionally, quantitatively measuring the qualitative factors influencing the buy or sell decisions helps speculators to better determine futures prices. This is important as they earn in the form of risk premium or the difference between the current and expected future price; with the latter being highly influenced by news items such as inventory decisions based on demand and supply (Gorton & Rouwenhorst, 2006). Thus, based on this study, it can be noted that traders of gold and silver futures can assess the risk premium, considering the short-term commodity price fluctuations caused by information like possible stock out, which is readily available, and take appropriate actions. However, to cater to the needs of the traders interested in any other commodity futures, more specific sources of information need to be considered.

Footnotes

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.

Funding

The authors received no financial support for the research, authorship and/or publication of this article.

References

Aggarwal

(2017). Exploring relation between Indian market sentiments and stock market returns. Asian Journal of Empirical Research, 7(7), 147–159. https://doi.org/10.18488/journal.1007/2017.7.7/1007.7.147.159

Alanyali

, Moat

H. S.

, & Preis

(2013). Quantifying the relationship between financial news and the stock market. Scientific Reports, 3(1), 3578. https://doi.org/10.1038/srep03578

Alfano

S. J.

, Feuerriegel

, & Neumann

(2015). Do pessimists move asset prices? Evidence from applying prospect theory to news sentiment. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.2602353

Altman

(2010). Prospect theory and behavioural finance. Behavioral Finance: Investors, Corporations, and Markets, 191–209. https://dx-doi-org.web.bisu.edu.cn/10.1002/9781118258415.ch11

Baker

H. K.

, & Nofsinger

J. R.

(2011). Behavioral finance: An overview. Behavioral finance: Investors, corporations, and markets, 1–21. https://doi.org/10.1002/9781118258415.ch1

Bloomfield

R. J.

(2010). Traditional vs. behavioral finance. Johnson School Research Paper Series, 22, 1–19.

Bognár

E. K.

(2016). Applying big data technologies in the financial sector–using sentiment analysis to identify correlations in the stock market. Computational Methods in Social Science, 4(1), 5–12.

Borovkova

(2015). The role of news in commodity markets. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.2587285

Chen

, Chong

T. T. L.

, & She

(2014). A principal component approach to measuring investor sentiment in China. Quantitative Finance, 14(4), 573–579. https://doi.org/10.1080/14697688.2013.869698

10.

Chen

, Zhao

, Li

, & Lu

(2020). A dynamic analysis of the relationship between investor sentiment and stock market realized volatility: Evidence from China. PLOS One, 15(12), e0243080. https://doi.org/10.1371/journal.pone.0243080

11.

Chevallier

, Gatumel

, & Ielpo

(2014). Commodity markets through the business cycle. Quantitative Finance, 14(9), 1597–1618. https://doi.org/10.1080/14697688.2013.842651

12.

Dash

S. R.

, & Mahakud

(2012). Investor sentiment, risk factors and stock return: Evidence from Indian non-financial companies. Journal of Indian Business Research, 4(3), 194–218. https://doi.org/10.1108/17554191211252699

13.

Dash

S. R.

, & Mahakud

(2013). Impact of investor sentiment on stock return: Evidence from India. Journal of Management Research, 13(3), 15.

14.

Deshmukh

B. G.

, Jain

P. S.

, Patwardhan

M. S.

, & Kulkarni

(2016). Spin-offs in Indian stock market owing to Twitter sentiments, commodity prices and analyst cecommendations. Proceedings of the International Conference on Advances in Information Communication Technology & Computing–AICTC, 16, 1–7. https://doi.org/10.1145/2979779.2979856

15.

Duxbury

(2015). Behavioral finance: Insights from experiments II: biases, moods and emotions. Review of Behavioural Finance, 7(2), 151–175. https://doi.org/10.1108/RBF-09-2015-0037

16.

Fan

, Zhao

, Chen

, & Xu

(2014). Anger is more influential than joy: Sentiment correlation in Weibo. PLoS One, 9(10), e110184. https://doi.org/10.1371/journal.pone.0110184

17.

Fang

, & Peress

(2009). Media coverage and the cross-section of stock returns. The Journal of Finance, 64(5), 2023–2052. https://doi.org/10.1111/j.1540-6261.2009.01493.x

18.

Ferguson

, Philip

, Lam

, & Guo

(2015). Media content and stock returns: The predictive power of press. Multinational Finance Journal, 19(1), 1–31. https://dx-doi-org.web.bisu.edu.cn/10.17578/19-1-1

19.

Gao

, Ren

, & Zhang

(2020). Googling investor sentiment around the world. Journal of Financial and Quantitative Analysis, 55(2), 549–580. https://doi.org/10.1017/S0022109019000061

20.

Gorton

, & Rouwenhorst

K. G.

(2006). Facts and fantasies about commodity futures. Financial Analysts Journal, 62(2), 47–68. https://doi.org/10.2469/faj.v62.n2.4083

21.

Gotthelf

, & Uhl

M. W.

(2019). News sentiment: A new yield curve factor. Journal of Behavioral Finance, 20(1), 31–41. https://doi.org/10.1080/15427560.2018.1432620

22.

Jin

, Yang

, & Liu

(2020). Stock closing price prediction based on sentiment analysis and LSTM. Neural Computing and Applications, 32(13), 9713–9729. https://doi.org/10.1007/s00521-019-04504-2

23.

Johnman

, Gepp

, & Vanstone

(2019). Harnessing investor sentiment using big data analytics. The Australasian Journal of Applied Finance; Sydney, 3, 3–8.

24.

Khedr

A. E.

, Salma

S. E.

, & Yaseen

(2017). Predicting stock market behaviour using data mining technique and news sentiment analysis. International Journal of Intelligent Systems and Applications, 9(7), 22–30. https://doi.org/10.5815/ijisa.2017.07.03

25.

, Liang

, Li

, Wang

, & Wu

D. D.

(2009). Network environment and financial risk using machine learning and sentiment analysis. Human and Ecological Risk Assessment: An International Journal, 15(2), 227–252. https://doi.org/10.1080/10807030902761056

26.

, Xie

, Chen

, Wang

, & Deng

(2014). News impact on stock price return via sentiment analysis. Knowledge-Based Systems, 69, 14–23. https://doi.org/10.1016/j.knosys.2014.04.022

27.

, Xu

, Yu

, & Tang

(2016). Forecasting oil price trends with sentiment of online news articles. Procedia Computer Science, 91, 1081–1087. https://doi.org/10.1016/j.procs.2016.07.157

28.

McGurk

, Nowak

, & Hall

J. C.

(2020). Stock returns and investor sentiment: Textual analysis and social media. Journal of Economics and Finance, 44(3), 458–485. https://doi.org/10.1007/s12197-019-09494-4

29.

S. Y. K.

, Liu

, & Yang

S. Y.

(2016). News sentiment to market impact and its feedback effect. Environment Systems and Decisions, 36(2), 158–166. https://doi.org/10.1007/s10669-016-9590-9

30.

Money Control (n.d.). Most active commodity on CX (Volume). Commodities: Most active. http://www.moneycontrol.com/commodities/mcx/most_active/all_0.html

31.

Multi Commodity Exchange India (n.d.). About MCX. https://www.mcxindia.com/about-us#:~:text=MCX%20is%20India’s%20leading%20commodity,futures%20contracts%20traded%20in%20FY20.

32.

Mukherjee

, & Goswami

(2017). The volatility of returns from commodity futures: Evidence from India. Financial Innovation, 3(1), 15. https://doi.org/10.1186/s40854-017-0066-9

33.

Peress

(2014). The media and the diffusion of information in financial markets: Evidence from newspaper strikes. The Journal of Finance, 69(5), 2007–2043. https://doi.org/10.1111/jofi.12179

34.

Rambocas

, & Pacheco

B. G.

(2018). Online sentiment analysis in marketing research: A review. Journal of Research in Interactive Marketing, 12(2). https://dx-doi-org.web.bisu.edu.cn/10.1108/JRIM-05-2017-0030

35.

Rao

, & Srivastava

(2013). Modeling movements in oil, gold, forex and market indices using search volume index and Twitter sentiments. Proceedings of the 5th Annual ACM Web Science Conference on–WebSci, 13, 336–345. https://doi.org/10.1145/2464464.2464521

36.

Research and Markets (2017). Global sentiment analysis software market 2017–2021. Research and Markets.

37.

Rupande

, Muguto

H. T.

, & Muzindutsi

P.-F.

(2019). Investor sentiment and stock return volatility: Evidence from the Johannesburg stock exchange. Cogent Economics & Finance, 7(1). https://doi.org/10.1080/23322039.2019.1600233

38.

Ryu

, Ryu

, & Yang

(2020). Investor sentiment, market competition, and financial crisis: Evidence from the Korean stock market. Emerging Markets Finance and Trade, 56(8), 1804–1816. https://doi.org/10.1080/1540496X.2019.1675152

39.

Soon

Y. C.

(2010). News which moves the market: Assessing the impact of published financial news on the stock market (Dissertations and Theses Collection [Open Access]). Singapore Management University.

40.

Shen

, Najand

, Dong

, & He

(2017). News and social media emotions in the commodity market. Review of Behavioral Finance, 9(2), 148–168. https://doi.org/10.1108/RBF-09-2016-0060

41.

, & Li

(2020). Dynamic sentiment spillovers among crude oil, gold, and Bitcoin markets: Evidence from time and frequency domain analyses. PLOS One, 15(12), e0242515. https://doi.org/10.1371/journal.pone.0242515

42.

Xia

, Yang

, Pan

, Zhang

, & An

(2020). Sentiment analysis for online reviews using conditional random fields and support vector machines. Electronic Commerce Research, 20(2), 343–360. https://doi.org/10.1007/s10660-019-09354-7

43.

Tabari

, Biswas

, Praneeth

, Seyeditabari

, Hadzikadic

, & Zadrozny

(2018). Causality analysis of Twitter sentiments and stock market returns. Proceedings of the First Workshop on Economics and Natural Language Processing (pp. 11–19). Association for Computational Linguistics. https://doi.org/10.18653/v1/W18-3102

44.

Tetlock

P. C.

(2007). Giving content to investor sentiment: The role of media in the stock market. The Journal of Finance, 62(3), 1139–1168. https://doi.org/10.1111/j.1540-6261.2007.01232.x

45.

Tetlock

P. C.

, Saar-Tsechansky

, & Macskassy

(2008). More than words: Quantifying language to measure firms’ fundamentals. The Journal of Finance, 63(3), 1437. https://doi.org/10.1111/j.1540-6261.2008.01362.x

46.

The Hindu Business Line (2020). MCX turnover rises 27% to ₹84 lakh crore in FY’20. https://www.thehindubusinessline.com/markets/commodities/mcx-turnover-rises-27-to-84-lakh-crore-in-fy20/article31282021.ece

47.

Tseng

K.-K.

, Lin

R. F.-Y.

, Zhou

, Kurniajaya

K. J.

, & Li

(2018). Price prediction of e-commerce products through internet sentiment analysis. Electronic Commerce Research, 18(1), 65–88. https://doi.org/10.1007/s10660-017-9272-9

48.

Uzar

, & Akkaya

G.cenk

(2013). The mental and behavioural mistakes investors make. International Journal of Business and Management Studies, 5(1), 120–128.

49.

Walliser

(2007). Cognitive economics. Springer Science & Business Media.

50.

Yadav

, & Vishwakarma

D. K.

(2020). Sentiment analysis using deep learning architectures: A review. Artificial Intelligence Review, 53(6), 4335–4385. https://doi.org/10.1007/s10462-019-09794-5

51.

Yelamanchili

R. K.

(2019). Impact of consumer sentiment on defensive and aggressive stock returns: Indian evidence. International Journal of Economics and Financial Issues, 9(4), 109–114. https://doi.org/10.32479/ijefi.8368

52.

Zhou

, Xu

, & Zhao

(2018). Tales of emotion and stock in China: Volatility, causality and prediction. World Wide Web, 21(4), 1093–1116. https://doi.org/10.1007/s11280-017-0495-4