3. Co-Citation Analysis of Commodity Prices
3.1. Cluster Analysis of Co-Cited Articles
If two articles appear together in the reference list of the third citing document, these two references form a co-citation relationship. If the two documents are cited by n documents together, the number of co-citations is n times. According to co-citation analysis, the degree of correlation between documents can be measured, and the collection of documents can be classified according to the degree of correlation, after which a cluster analysis network diagram of professional papers in that discipline can be generated. In Figure 8, showing the co-citation network of sample articles, node size in the graph indicates the frequency of citations. The higher the frequency of citations, the larger the nodes. The connection between two nodes designates these two documents have been cited together. Indicating a change in the research direction, the color of the infographic changed from dark to light over the 1990 to 2020 period, and the research focus expanded from an agricultural commodity price and oil price dynamic to an oil price prediction and financialization.
Figure 8. Clusters of co-cited articles.
In
Table 4, which depicts a summary of each cluster’s details, cluster size represents the number of articles it includes. Silhouette is a parameter proposed by Kaufman and Rousseeuw in 1990 to evaluate the effect of clustering through network homogeneity
[15]. The Silhouette value lies between 0–1, and the closer it is to 1, the higher the homogeneity of the network and the higher the credibility of clustering. The publication year is the average publication time of articles in this cluster. According to
Table 4, the largest clusters are Cluster #0, #1, #2, #3, and #4. Their articles focus on the exploration of the correlation between commodity prices and the stock market, the exchange rate, oil price uncertainty, and crude oil price forecasting. In addition, most clusters’ Silhouette values are above 0.7, indicating that these clusters are highly homogeneous and highly reliable.
Table 4. Citation clustering network.
Cluster ID |
Cluster Label |
Cluster Size |
Silhouette |
Year (Average) |
#0 |
stock market |
209 |
0.656 |
2006 |
#1 |
exchange rate |
141 |
0.698 |
2013 |
#2 |
crude oil price forecasting |
126 |
0.729 |
2012 |
#3 |
oil price uncertainty |
103 |
0.749 |
2014 |
#4 |
agricultural commodity price |
76 |
0.854 |
2009 |
#5 |
Singapore economy |
70 |
0.954 |
1996 |
#6 |
agricultural employment |
47 |
0.995 |
1991 |
#7 |
EMD-based neural network ensemble |
47 |
0.969 |
2001 |
#9 |
long-run price |
38 |
1 |
1993 |
#20 |
high commodity price |
7 |
0.996 |
2008 |
#26 |
high oil price |
4 |
0.996 |
2003 |
The citation network and cluster analysis show that the research hotspots of commodity prices focus on influencing factors, impacts on macroeconomy, price forecasts, and commodity financialization; moreover, these hotspots form a research network centered on oil prices. The detailed discussion is shown in Table 5:
- (1)
-
Influencing factors of commodity prices (Cluster #1, #9, #20, #26)
The influencing factors of commodity price fluctuations have been widely focused on by many scholars. The contradiction between supply–demand and inventory change are generally considered the basic factors affecting international commodity price fluctuations. Since the operation of the global economy is fundamental to commodity supply and demand, but financial markets can have a spillover effect into the commodity market, the coordinated change of commodity prices becomes a goalpost for planners. Speculative factors have pushed up the prices of major commodities in the past, making the strong demand from emerging economies another reason for the overall rise in commodity prices. Factors including natural disasters, cyclical changes in production, and policy adjustments also considerably impact commodity prices. Differences in national economic systems also lead to differences in commodity prices between countries. The economic system affects the speed and strength of the transmission of international commodity prices to domestic commodity prices, and ultimately reflects in domestic prices.
A substantial increase or decrease in commodity prices frequently results from a combination of multiple factors. Hamilton
[16] studied some factors leading to changes in crude oil prices, including supply and demand, commodity speculation, monopolistic oligarchs, and resource depletion. Mueller et al.
[17], analyzing the reasons why food prices doubled in the world from March 2007 to March 2008, found multiple factors, including greater energy and fertilizer costs, increased financial speculation in commodity markets, export restrictions, dollar depreciation, biofuel production increase, poor harvests, reduced world food reserves, increased foreign exchange reserves, and the continuing increase in world population and expectations of affluence.
Academic circles have continuously analyzed the factors affecting commodity price fluctuations. Ranging from the basics of supply and demand to the impact of financial speculation and analysis of interruptive events, and from linear to nonlinear, the research in these directions continues to deepen comprehensively
[18][19][20][21][22][23][24][25].
- (2)
-
The impact of commodity price fluctuations on macro economy (Cluster #3, #4, #5, #6)
The impact of commodity price fluctuations on the macroeconomy is another noteworthy research direction that has developed to a great extent.
Building on the foundational work of Hamilton
[26] four decades ago, scholars have segued to study the oil price–economic growth relationship
[27][28][29][30] or taken up nonlinear perspective considerations
[31][32][33]. Important in international trade, the relationship between commodity prices and exchange rates has also received key attention. Chen
[34] used monthly panel data from G7 countries to study the long-term relationship between oil prices and real exchange rates, and found that a co-integration relationship between them and oil prices can better predict exchange rate trends. Other literature uses models such as the vector autoregressive model (VAR) and generalized autoregressive conditional heteroskedasticity model (GARCH) to select different countries as research objects to explore the relationship between commodity prices and exchange rates
[35][36][37][38][39][40][41][42].
The unemployment rate and consumer price (CPI) are also variables of concern. Papapetrou
[43] used the VAR model to explore the relationship between oil prices and several macroeconomic variables in Greece, and found that changes in oil prices affect actual economic activities and employment. Ewing and Thompson
[44] used Hodrick–Prescott and Baxter–King filters, and a full-sample asymmetric Christiano–Fitzgerald band-pass filter, to study the cyclical linkage between crude oil prices and production, consumer prices, unemployment, and stock prices. Guerrero-Escobar et al.
[45] used 17 highly heterogeneous countries as samples to study responses of industrial activity, inflation, interest rates, and exchange rates to oil price shocks.
The impact of commodity prices on sustainable development has received increasing attention in recent years, mainly energy prices. Borzuei et al.
[8] used a threshold model to study the impact of energy prices on Iran’s sustainable development under different economic growth regimes, which is mainly achieved by promoting the use of renewable energy. Similarly, Li and Leung
[46], Mukhtarov et al.
[47], Karacan et al.
[48], and Mukhtarov et al.
[49] confirmed the role of energy prices on renewable energy consumption, and further affects the achievement of sustainable development goals. Umar et al.
[50] found that effective energy price policies can reduce carbon emissions and achieve SDGs; Ma et al.
[51], Sha et al.
[52], Lee and Chong
[53], Guo et al.
[54], Malik et al.
[55], and Ike et al.
[56] also confirmed the same point.
- (3)
-
Commodity price forecast (Cluster #2, #7)
The drastic fluctuations in commodity prices have attracted many scholars to study commodity price forecasting. When the prices of commodities are relatively stable, forecasting pays more attention to the general trend of prices and the overall price range, that is, the first-order moment of prices. But when commodity prices fluctuate wildly, forecasting needs to focus on the degree of price volatility, that is, the second-order moment of prices.
At present, there are mainly three methods to forecast commodity prices: (1) traditional econometric models, such as the autoregressive moving average model (ARMA), autoregressive integrated moving average model (ARIMA), and generalized autoregressive conditional heteroskedasticity model (GARCH); (2) machine learning methods, such as support vector machines (SVM), artificial neural networks (ANN), and Random Forest (RF); and (3) hybrid models combining the above two methods
[57]. Traditional econometric models are based on strict linear assumptions and cannot introduce too many independent variables. Machine learning methods, because they can effectively distinguish random factors and capture hidden nonlinear characteristics that traditional econometric models fail to do, can introduce more independent variables to maximize the control of influencing factors.
For the traditional econometric models, some scholars have proposed improved models for commodity price prediction. Herrera et al.
[58] used high-frequency intra-day realized volatility data to evaluate the relative prediction performance of various models, including risk measurement models, the GARCH model, asymmetric GARCH model, split-integral GARCH model, and Markov transformed GARCH model. Gupta and Wohar
[59] used a qualitative vector autoregressive to predict monthly returns of oil and stocks. Based on the GARCH-X model, Gavriilidis et al.
[60] used oil shocks of unknown sources as exogenous variables to study their impact on the accuracy of spot price predictions.
For machine learning methods, scholars have attempted to decompose the sequence through singular spectrum analysis (SSA) or empirical mode decomposition (EMD), predict the decomposed sequence, and finally synthesize the prediction of the original sequence. Yu and Wang
[61] proposed a neural network ensemble learning paradigm based on EMD for the forecast of world crude oil spot prices. Wang and Li
[62] proposed SSA-NN, a neural network prediction method based on SSA smoothing, and used four criteria to compare the performance of this model with the basic neural network model. Their results show that the SSA-NN model is superior to the basic neural network model in terms of performance and accuracy to capture trend changes.
More scholars have focused on innovating on basic machine learning methods to improve prediction accuracy. For example, Huang and Wu
[57] used deep multi-core learning (DMKL) to predict oil prices, verified that the model was superior to the traditional model through real data, and significantly reduced prediction errors. Zhang and Na
[63] proposed an agricultural product price prediction model that combines fuzzy information granulation, a mind evolutionary algorithm (MEA), and a support vector machine (SVM). Their experiments proved that the model has a higher prediction accuracy and faster calculation speed.
In the hybrid model, many models combine the advantages of measurement models and machine learning to improve prediction accuracy. Zhang et al.
[64] proposed the EEMD–PSO–LSSVM–GARCH hybrid model to predict oil prices. Kristjanpoller and Minutolo
[65] used the ANN-GARCH model to predict oil spot and future prices, and successfully improved that prediction model in terms of volatility and spot prices.
- (4)
-
Financialization of commodities (Cluster #0)
As a foundational study on the relationship between commodity prices and financial markets, Sadorsky
[66] adopted the VAR model that incorporated monthly data from January 1947 to April 1996 in the United States. It was found that the shocks of oil prices would significantly affect the actual returns of stocks, and this effect was more pronounced in 1986. Afterward, scholars have chosen various research objects and different time periods to analyze the impact of commodity price changes on stock price changes or stock returns
[43][67][68][69][70][71][72][73].
The 2006–2008 food crisis and the 2008 global financial crisis caused violent fluctuations in commodity prices. This led the academic community to conclude that basic supply and demand factors alone could not explain drastic changes in prices in a short period of time, sensing some of the change could be attributed to financial factors, such as the financialization of commodities, that is, significant index investment flows into commodity markets. Morana
[74] found that since 2000, financial shocks have played an important role in boosting oil prices, and this effect has become more pronounced after 2008. Tilton
[75] studied the impact of investor demand on commodity spot prices and predicted that, if the futures market is strong, a surge in investor demand will increase the price of the futures market and directly affect spot market price. However, when the futures market is relatively weak, investor behavior would have little impact on spot prices. The financialization of commodity markets has also resulted in excessive linkages between commodity prices
[76].