CHINA’S IMPORT TARIFFS ON U.S. SOYBEAN EXPORTS 2018-2022: EFFECTS ON INFORMATION TRANSFER BETWEEN MARKETS IN CHINA, THE U.S. AND BRAZIL A Thesis Presented to the Faculty of the Graduate School of Cornell University in Partial Fulfillment of the Requirements for the Degree of Master of Science by Juraj Stelmach August 2023 © 2023 Juraj Stelmach ALL RIGHTS RESERVED ABSTRACT This paper investigates the impacts of China’s import tariffs on U.S. soybean exports in mid-2018 on price diffusion and information transfer relationships between futures prices in the U.S., China, and Brazil. To assess the informa- tion transfer relationships, we use the Reduced Vector Autoregression (RVAR) model to generate pairwise tests on the causal covariate differences between the U.S., China, and Brazilian soybean futures prices before and after mid-2018. Results show that previous patterns of price signaling between the U.S. closing and Chinese opening soybean futures prices, and the Chinese closing and U.S. opening soybean futures prices, have all but evaporated since the import tariffs imposed on U.S. exports to China in mid-2018, and recovered only after January 2020. This import tariff provided a natural experiment on the effects of a tariff on price information transfers between different markets worldwide as global trade patterns in soybeans changed. BIOGRAPHICAL SKETCH Personal Information • Name: Juraj Stelmach • Nationality: Slovak • Hometown: Bratislava, Slovakia Education • Bachelor of International Economics, University of British Columbia (UBC), 2018 • Master of Science in Applied Economics & Management, Cornell Univer- sity, 2023 Professional Experience • Research Assistant, UBC, 2018 • Risk Consultant, KPMG, 2018 • Budget Analyst, North Atlantic Treaty Organization (NATO), 2019 • Income Analyst, Scotiabank, 2019-2021 • Graduate Research Assistant, Cornell University, 2022-2023 Research or Specialization • International Trade iii • Futures Markets • Commodity Prices and Biofuels Personal Interests • Reading Eastern European fiction and science fiction novels • Reading historical nonfiction • Ancient Philosophy • Wine tasting • Scuba Diving • Sailing • Martial Arts • Firearm Safety iv This thesis is dedicated to my parents, who have always supported and encouraged me throughout my academic journey. v ACKNOWLEDGEMENTS I would like to express my heartfelt gratitude to several individuals who played a crucial role in shaping my academic journey and making this article possible. First and foremost, I am deeply indebted to Dr. Ries, Dr. Kotwal, and Dr. Alviarez for their unwavering support and mentorship in the field of interna- tional trade. Their guidance and passion for the subject sparked my own inter- est and set me on the path to pursue a career in economics. I will forever be grateful for their profound impact on my academic and professional growth. I would like to extend my thanks to Dr. McIntyre, whose thought-provoking lectures on economic models and global macroeconomics deepened my under- standing and broadened my perspective on macroecnomics and financial mar- kets, which played a fundamental role in understanding how the world works. My sincere appreciation also goes to Dr. Alviarez for teaching me the intri- cacies of academic research and providing me with invaluable insights into the world of economics. Additionally, I am grateful to the BIE program for provid- ing me with a comprehensive understanding of various economic concepts and principles. I must also acknowledge Mr. MacRury, whose captivating classes ignited my passion for economics and motivated me to delve deeper into the subject. I am immensely thankful to Dr. de Gorter and Dr. Turvey for their support and encouragement throughout my journey in applied economics. Their guid- ance helped me solidify my understanding and inspired me to pursue research in the field. Special thanks to my parents who believe that education is the best invest- ment in oneself, providing the knowledge, skills, and opportunities needed to unlock one’s potential, broaden horizons, and achieve personal growth and suc- vi cess. I am grateful for my girlfriend Victoria, although she may have given up af- ter reading the fifth page, her dedication shone through when she attended my defense at 5 in the morning (AZ time), despite being sick from food poisoning. Lastly, I cannot forget to acknowledge the countless cups of coffee that kept me awake during those long nights of writing this article. Their magical powers transformed me into a caffeinated writing machine, and I am truly indebted to them for helping me meet deadlines and complete this work. To all the individuals mentioned above, I say, thank you from the bottom of my heart for being instrumental in this achievement. Your support and mentor- ship have been invaluable, and I am grateful beyond words. vii TABLE OF CONTENTS Biographical Sketch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii 1 Introduction 1 2 Literature Review 6 3 Background 11 4 Country Specific Soybeans Production 14 4.1 U.S. Soybeans . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 4.2 Brazilian Soybeans . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 5 International Trade of Soybeans & China 21 6 The Economic Framework 26 6.1 Trade Theory and Comparative Advantage . . . . . . . . . . . . . 26 6.2 Tariff and World Prices . . . . . . . . . . . . . . . . . . . . . . . . . 28 6.3 Empirical Evidence on U.S. and Brazilian Spot Prices . . . . . . . 31 7 Trade War and Grains Storage 33 7.1 Soybean Futures Nonconvergence . . . . . . . . . . . . . . . . . . 35 7.2 Corn Futures Nonconvergence . . . . . . . . . . . . . . . . . . . . 37 7.3 Economics of Nonconvergence . . . . . . . . . . . . . . . . . . . . 37 8 Data 40 8.1 Price Discovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 8.2 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 42 9 Methodology 45 9.1 RVAR Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 9.2 Granger Causality Tests . . . . . . . . . . . . . . . . . . . . . . . . 47 10 Stationarity 49 10.1 Autocorrelation Function (ACF) and Partial Autocorrelation Function (PACF) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 10.2 Augmented Dickey-Fuller (ADF) Test . . . . . . . . . . . . . . . . 52 10.3 Kwiatkowski-Phillip-Schmidt-Shin test (KPSS test) . . . . . . . . . 57 viii 11 Results 59 11.1 Using First Order Differences (FOD) Model in log form . . . . . . 59 11.2 Signaling between CME and DCE markets . . . . . . . . . . . . . 59 11.2.1 Prior TW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 11.2.2 TW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 11.2.3 Post TW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 11.3 Signaling between B3 and DCE markets . . . . . . . . . . . . . . . 65 11.3.1 Prior TW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 11.3.2 TW + Post TW Periods . . . . . . . . . . . . . . . . . . . . . 67 12 Robustness 70 12.1 Level-level specification . . . . . . . . . . . . . . . . . . . . . . . . 71 12.2 Signaling between CME and DCE markets . . . . . . . . . . . . . 71 12.2.1 Prior TW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 12.2.2 TW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 12.2.3 Post TW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 12.2.4 After adjustment for the pandemic year . . . . . . . . . . . 74 12.3 Signaling between B3 and DCE markets . . . . . . . . . . . . . . . 75 12.3.1 Prior TW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 12.3.2 TW + Post TW Periods . . . . . . . . . . . . . . . . . . . . . 76 12.4 Complete Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 12.5 Signaling between CME and DCE markets . . . . . . . . . . . . . 78 12.5.1 Prior TW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 12.5.2 TW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 12.5.3 Post TW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 12.5.4 Adjusted for the pandemic year . . . . . . . . . . . . . . . 81 12.6 Signaling between B3 and DCE markets . . . . . . . . . . . . . . . 82 12.6.1 Prior TW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 12.6.2 TW + Post TW Periods . . . . . . . . . . . . . . . . . . . . . 83 13 Conclusion 86 14 Bibliography 88 ix LIST OF TABLES 8.1 Data Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 8.2 Summary statistics and unit root tests for First Order Differenced Data Series in log form. . . . . . . . . . . . . . . . . . . . . . . . . 44 9.1 Granger Causality Test: Model 1 vs. Model 2 . . . . . . . . . . . . 48 10.1 ADF Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 10.2 ADF Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 10.3 KPSS Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 10.4 Estimation results for equation DCE.Open . . . . . . . . . . . . . 58 11.1 Estimation results for equation DCE.Open . . . . . . . . . . . . . 59 11.2 Estimation results for equation CME.Open . . . . . . . . . . . . . 60 11.3 Estimation results for equation DCE.Open . . . . . . . . . . . . . 61 11.4 Estimation results for equation CME.Open . . . . . . . . . . . . . 61 11.5 Estimation results for equation DCE.Open . . . . . . . . . . . . . 62 11.6 Estimation results for equation CME.Open . . . . . . . . . . . . . 62 11.7 Estimation results for equation DCE.Open . . . . . . . . . . . . . 63 11.8 Estimation results for equation CME.Open . . . . . . . . . . . . . 64 11.9 Estimation results for equation DCE.Open . . . . . . . . . . . . . 66 11.10 Estimation results for equation BRZ.Open . . . . . . . . . . . . . 66 11.11 Estimation results for equation DCE.Open . . . . . . . . . . . . . 67 11.12 Estimation results for equation BRZ.Open . . . . . . . . . . . . . 68 12.1 Estimation results for equation DCE.Open . . . . . . . . . . . . . 71 12.2 Estimation results for equation CME.Open . . . . . . . . . . . . . 71 12.3 Estimation results for equation DCE.Open . . . . . . . . . . . . . 72 12.4 Estimation results for equation CME.Open . . . . . . . . . . . . . 72 12.5 Estimation results for equation DCE.Open . . . . . . . . . . . . . 73 12.6 Estimation results for equation CME.Open . . . . . . . . . . . . . 73 12.7 Estimation results for equation DCE.Open . . . . . . . . . . . . . 74 12.8 Estimation results for equation CME.Open . . . . . . . . . . . . . 74 12.9 Estimation results for equation DCE.Open . . . . . . . . . . . . . 75 12.10 Estimation results for equation BRZ.Open . . . . . . . . . . . . . 76 12.11 Estimation results for equation DCE.Open . . . . . . . . . . . . . 76 12.12 Estimation results for equation BRZ.Open . . . . . . . . . . . . . 77 12.13 Estimation results for equation DCE.Open . . . . . . . . . . . . . 78 12.14 Estimation results for equation CME.Open . . . . . . . . . . . . . 78 12.15 Estimation results for equation DCE.Open . . . . . . . . . . . . . 79 12.16 Estimation results for equation CME.Open . . . . . . . . . . . . . 79 12.17 Estimation results for equation DCE.Open . . . . . . . . . . . . . 80 12.18 Estimation results for equation CME.Open . . . . . . . . . . . . . 80 12.19 Estimation results for equation DCE.Open . . . . . . . . . . . . . 81 x 12.20 Estimation results for equation CME.Open . . . . . . . . . . . . . 81 12.21 Estimation results for equation DCE.Open . . . . . . . . . . . . . 82 12.22 Estimation results for equation BRZ.Open . . . . . . . . . . . . . 83 12.23 Estimation results for equation DCE.Open . . . . . . . . . . . . . 83 12.24 Estimation results for equation BRZ.Open . . . . . . . . . . . . . 84 xi LIST OF FIGURES 4.1 Soybeans 2022 Production by County1 . . . . . . . . . . . . . . . 17 4.2 Brazilian Soybeans Production2 . . . . . . . . . . . . . . . . . . . 19 4.3 Brazil’s major road network in 2017 and BR-1633 . . . . . . . . . 20 4.4 Brazil’s Transportation Costs and Soybean Prices4 . . . . . . . . . 20 5.1 Chinese Monthly Soybean Imports . . . . . . . . . . . . . . . . . . 22 5.2 Chinese Monthly Soybeans Consumption, Production and Imports 24 5.3 ASF Outbreaks in China5 . . . . . . . . . . . . . . . . . . . . . . . 24 6.1 U.S. Trade Balance from 1989 to 20226 . . . . . . . . . . . . . . . . 27 6.2 The effect of an ad valorem import tariff on U.S. and China’s Soybean Market . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 6.3 Pre-import tariff, Brazil’s export are large enough to fill the void if China reduces demand for U.S. soybeans . . . . . . . . . . . . . 30 6.4 U.S. and Chinese Spot Prices . . . . . . . . . . . . . . . . . . . . . 31 7.1 U.S. Soybeans Nonconvergence . . . . . . . . . . . . . . . . . . . 35 7.2 Source: Bloomberg . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 7.3 Convergence under Normal Inventory Demand7 . . . . . . . . . 38 7.4 Nonconvergence under Excess Inventory Demand8 . . . . . . . . 39 8.1 Time Series and Trade War Periods . . . . . . . . . . . . . . . . . 44 10.1 CME ACF and PACF . . . . . . . . . . . . . . . . . . . . . . . . . . 50 10.2 DCE ACF and PACF . . . . . . . . . . . . . . . . . . . . . . . . . . 51 10.3 B3 ACF and PACF . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 10.4 Time Series for Opening and Closing Soybean Prices for CME, DCE and B3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 10.5 FOD Time Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 11.1 P-Values for CME and DCE signaling . . . . . . . . . . . . . . . . 65 11.2 P-Values for B3 and DCE signaling . . . . . . . . . . . . . . . . . . 69 12.1 P-Values for CME and DCE signaling . . . . . . . . . . . . . . . . 75 12.2 P-Values for B3 and DCE signaling . . . . . . . . . . . . . . . . . . 77 12.3 P-Values for CME and DCE signaling . . . . . . . . . . . . . . . . 82 12.4 P-Values for B3 and DCE signaling . . . . . . . . . . . . . . . . . . 84 xii CHAPTER 1 INTRODUCTION In recent years, the geopolitical climate started to shift from globalization to pro- tectionism. According to the Ricardian model, globalization and international trade increases wealth and consumer surplus in both trading countries. This was mostly seen through offshoring parts of production. For example, Boeing, a U.S. leading commercial airplane manufacturer, partners with 58 countries to produce its aircrafts. This is an astonishing number of countries that one manufacturer is dependent on. Our case study began with the Trump election which occurred in November of 2016. President Trump aimed to reduce the trade deficit with China, which refers to the difference in imports and exports between the two nations. He expressed discontent with China buying fewer goods from the U.S. than the U.S. buying from China, which resulted in a gap valued at $375bn in 2017. Moreover, he has made allegations against Chinese companies for intellectual property theft from U.S. companies and demanded that Beijing make changes to its regulations. Trump’s political campaign benefited from people’s discontent with current terms in international trade. Not between the countries, but between the corpo- rations and citizens. According to Globalization and Its Discontents by Joseph Stiglitz, as an example he provided the proposed TPP trade agreement between the U.S. and Asia. Totaling 40 percent of global economy, the estimated GDP growth in the U.S. from this agreement over the next 15 years was estimated to be only about 0.15 percent, which is a very negligible amount. A similar agree- ment made during the Obama administration mostly helped Multinational Cor- porations and not average citizens. Many argue that the aim of this policy was 1 to increase the U.S. sphere of influence. Trump’s goal to rejuvenate manufac- turing could have not been successful, because only 8 percent of the U.S. labor force works in that industry. However, it helped to solidify his political base under the same ideology, and turned his focus from corporations to a country, China. The globalization movement in the 1990s offshored blue-collar jobs in both developed and developing countries, including China. This posed an is- sue as the communities in the developed countries that had experienced a loss in international trade were not compensated, thus left behind in the process. This led to the Trump administration proposing 25 percent tariffs on $34 billion Chinese imports in April 2018, including materials such as steel, aluminium, solar panels, and washing machines. In retaliation, China responded with tar- iffs on July 6th with a 25 percent tariff on $50 billion of U.S. goods, including soybeans. Tariffs are customs duties levied by a country or economic bloc on imported goods from a trading partner. The purpose of this is to increase the price of those goods to make them less appealing to consumers than domestic goods. The tariffs are paid by purchasers of the goods when they cross the bor- der into the country that has imposed the tariffs. To put this into perspective, China is the largest global soybeans importer consisting of 60 percent of global soybean imports. The U.S. on the other hand, is one of the largest global ex- porters of soybeans along with Brazil. In 2017, the year prior to trade war, the U.S. exported almost 50 percent of its harvest and 54.6 percent of total soybeans exports went to China, which is 32.9 million tons. The tariffs had a ripple effect on U.S. farmers, grain storages, and futures markets. This paper investigates the impacts of China’s import tariff on U.S. soybean exports starting on July 6th 2018 on price diffusion and information transfer relationships between prices of soybean futures contracts traded on the Chicago Mercantile Exchange (CME), 2 Dalian Commodity Exchange (DCE), and Brazilian Mercantile and Futures Ex- change (B3). In the context of price fluctuations resulting from the Brownian motion pro- cesses, price changes occur in response to new sources of information that be- come a part of the overall global set of information. Although some players may not be aware of the sources of information, the signals from those who possess this knowledge are enough to influence the market. For instance, the opening price in China is determined by the flow of information from changes in U.S. and Brazilian markets, which in turn reflect the current expectations of supply and demand in the Chicago and Brazilian futures markets. A rise in prices in the U.S. or Brazil can indicate an increase in demand or a decrease in supply, or both, and this information is then transmitted and incorporated into the infor- mation set used by Chinese traders to determine their own supply and demand position. As a result, U.S., Brazilian, and Chinese markets are co-integrated. The main question is whether the flow of information and cross-market sig- naling that we examine below existed before the imposition of import tariffs on soybeans, and whether it remained continuous or became disrupted afterwards. Since 2012, China has emerged as the dominant market for U.S. soybean ex- ports, with price relationships between U.S. and Chinese futures prices traded at CME and DCE in Liaoning Province being crucial to our analysis. The impo- sition of import tariffs on U.S. soybean exports to China has disrupted the tradi- tional supply-demand information flows in global trade, leading to distortions, uncertainty, and ambiguity and disrupting transparent price discovery. The U.S. and China soybean markets were once strongly interconnected, with significant research showing the relationship between their futures and spot prices. Fung 3 et al. (2010) and Hua and Chen (2007) demonstrated a significant relationship between U.S. and Shanghai futures markets for copper, aluminum, soybeans, and wheat, while Fung et al. (2003) proved that the U.S. soybean futures mar- ket played a dominant role in transmitting trading information to the Chinese market, although this role may be diminishing as China’s market power and internal pricing system matures. While there is a significant interdependence between Chinese and U.S. mar- kets for commodities such as soybeans, wheat, corn, and sugar, it is widely believed that China’s growing economic power and dependence on U.S. mar- kets pose a greater risk of tension than not (Gale, 2015; Jiang, 2016). However, this belief is being challenged due to the ongoing trade war. Despite this, sev- eral U.S. studies have confirmed that prior to the imposition of import tariffs on soybeans, U.S. soybean futures’ overnight return and Chinese No.1 soybean futures’ daytime return had a simultaneous effect on each other (Li and Hayes, 2017; Han et al. 2013). Our study aligns closely with Li and Hayes’ (2017) investigation into the cointegration of soybean futures prices between the U.S., China, and Brazil, but differs in two significant ways. Firstly, our study has a more narrow scope, focusing only on short-term pricing of U.S. and Chinese soybean futures be- fore and after the imposition of the import tariff in 2018. Second, we measure the basis of price changes, focusing on the relationship between U.S., Brazilian, and Chinese prices, rather than their changes over time, as examined by Li and Hayes. Specifically, we examine the relationship between the Chinese opening price against the previous U.S. closing price (or B3 closing price) and the U.S. opening price (or B3 opening price) against the previous Chinese closing price, 4 solely focusing on CME, DCE, and B3 trades and their impact on Chinese spot price. While overnight trading occurs on GLOBEX exchange in all markets, these are less liquid, and we use DCE #1 soybean futures contracts, which are priced to non-GM soybeans, as opposed to the DCE #2 contracts, which are a mixed blend. We investigate the impact of the import tariff on Chinese soybean spot prices. In June 2018, when China announced the tariff on soybean imports, the U.S. had already planted its soybean crops. Typically, after harvest, exports would deplete stored soybeans in elevators, which would then be replenished by the next harvest, thus maintaining a storage equilibrium. However, due to the increased tariffs and import embargo, many U.S. areas struggled to find off- farm storage, resulting in the 2018 soybean harvest flooding the cash market. This widened the basis across the U.S. and caused a drop in spot prices. In con- trast, China was able to replenish its storage with soybean imports from Brazil and Argentina, as well as maintain its basis, ultimately stabilizing prices. Since CME & DCE and B3 & DCE operate in different time zones, the clos- ing prices of one market may impact the opening prices of the other. This pa- per uses the well-established Reduced Vector Autoregressive (RVAR) model to investigate signaling effects of opening and closing prices of soybean futures prices. Our approach enables us to identify causal relationships between soy- bean futures prices in the U.S., China, and Brazil. By applying the RVAR model, we evaluate the differences in causal covariates of soybean futures prices before and after the implementation of import tariffs on July 6th, 2018. 5 CHAPTER 2 LITERATURE REVIEW One of the earliest works on commodities pricing has been conducted by Gus- tav Cassel in The Theory of Social Economy. Cassel (1923) studied the concept of purchasing power parity (PPP) and concluded that exchange rates adjust to reflect a country’s PPP, meaning that two goods should be priced the same when measured in a common currency. This idea gave rise to the Law of One Price (LOOP), which states that two goods should have the same price in different markets once we account for barriers to trade and transportation costs. Despite this, there are many frictions between trading countries that may affect prices, such as production costs and government policies. Thus, exchange rates may not always capture these differences, and prices may not be equal across coun- tries. In this analysis, the focus is on the information transfer from CME, DCE, and B3 for soybeans, which are theoretically the same but may have nutritional differences in different countries that add to the friction. But our main interest is the friction of tariffs on U.S. soybeans and its impact on price signalling. Fama (1970) put forth the concept of efficient markets, that all available infor- mation is already incorporated into the price. This means that financial markets are highly efficient in reflecting all publicly available information, making it im- possible for anyone to consistently outperform the market. This concept applies to our case of soybean futures markets as well, where hedgers, investors, and speculators are also unable to consistently outperform the market. Therefore, futures prices in different countries should already reflect all available informa- tion. The study by Bigman, Goldfarb, and Schechtman (1983) examines the rela- 6 tionship between futures market efficiency and the time content of the informa- tion sets. They argue that the degree of market efficiency depends on the infor- mational content of futures prices and the timing of their formation relative to the arrival of new information. The authors focus on the time series properties of futures prices for soybeans, corn, and wheat. They use a statistical model to examine the degree to which past prices, lagged by various time intervals, are able to predict future prices. Their findings reveal that the predictive power of past prices declines rapidly as the time interval increases, indicating that futures prices are highly responsive to new information. Fung, Leung, and Xu (2003) discuss the information flows between the United States and China in commodity futures trading. The study suggests that the United States has a stronger impact on Chinese commodity futures markets of soybeans and copper. The authors note that these findings can have impor- tant implications for market participants, including traders and policymakers, as well as for risk management and pricing strategies. Additionally, the au- thors suggest that increased transparency and communication between the two countries could potentially enhance market efficiency and stability. Fung, Liu, and Tse (2010) examine the information flow and market effi- ciency between the U.S. and Chinese aluminum and copper futures markets us- ing the Vector Error Correction Model (VECM). Their study finds that both U.S. and Chinese markets are highly cointegrated and both are efficient at reflecting new market information through price. Fung, Tse, Yau, and Zhao (2013) examine the development and character- istics of China’s commodity futures market, analyzing its growth, trading vol- ume, market structure, and regulatory environment. They investigate the fac- 7 tors that contribute to the market’s potential to become a leading player in the world commodity futures markets. Their results show that China is becoming a market leader in many commodities and efficiently reflect all the prices with minimal lead-lag relationships with foreign markets. However, soybeans came as an exception in their results for both return models, which are open-close and close-close. This is encouraging for our current model specification. But we are now cautious of the fact that p-values for a close-close model may be higher than for open-close models with the incorporation of overnight prices. Han, Liang and Tang (2013) study the cross-market soybean futures price discovery, which refers to the process by which futures prices for soybeans are determined in different markets, such as the Dalian Commodity Exchange (DCE) in China and the Chicago Board of Trade (CBOT) in the United States. They applied both the Structural Vector Autoregressive Model (SVAR) and VECM. The question at hand is whether the DCE affects the CBOT. Their re- sults found that the DCE has become increasingly important in global soybean price discovery, with its influence on the CBOT growing over time. This is due to China’s growing role as a major importer of soybeans, which has led to in- creased trading activity on the DCE and greater demand for price information. The DCE has been found to have a significant impact on the CBOT during pe- riods of high volatility and uncertainty in global soybean markets, with price movements on the DCE being transmitted to the CBOT having similar infor- mation magnitude as signaling from CBOT to DCE. Therefore, the authors con- clude that DCE is not a satellite market but has a prominent role in the global soybean futures price discovery. Li and Lu (2012) study analyzes the cross-correlations between CBOT corn 8 and soybean markets and the DCE corn and soybean markets. The analysis covers the period from January 2013 to December 2017. The study finds that the CBOT and DCE futures markets are positively correlated, indicating a strong interdependence between the markets. The study also finds that the soybean markets are more closely linked than the corn markets, which may be due to the larger role that soybeans play in international trade. Li and Hayes (2017) employ a threshold cointegration approach to exam- ine the relationship between the Chicago Board of Trade (CBOT), the Dalian Commodity Exchange (DCE), and the Brazil Mercantile and Futures Exchange (BM&F) soybean futures markets. Their results confirm that the U.S. is the lead- ing soybean futures market, which price influences other markets in China and Brazil. However, they also found that the unidirectional link between the U.S. and China or U.S. and Brazil has been weakened, and the opposite direction of price causality recently appeared. Liu and An (2011) examines the research related to information transmission between the U.S. and Chinese commodity futures markets. The authors investi- gate the extent to which information flows between these markets, providing evidence of the interconnectedness and information linkages between them. By using both synchronous and non-synchronous trading information from Chinese futures/spot markets, the New York Mercantile Exchange (NYMEX), Chicago Board of Trade (CBOT), and CME Globex futures markets for cop- per and soybeans, the authors show that there is a bidirectional relationship in terms of price and volatility spillovers between U.S. and Chinese markets, with a stronger effect from U.S. to Chinese markets than the other way around. How- ever, they also highlight the increasing prominence of Chinese futures markets 9 on the world markets. Shrestha, K., Subramaniam, R., and Thiyagarajan, T. (2020) argue that the price discovery happens in the futures markets with the exception for cocoa, for which the price discovery happens in the spot markets. The authors examine seven commodities: soybean oil, soybeans, soybean meal, corn, wheat, cocoa, and coffee. The findings help us to specify our model to primarily focus on the futures prices, rather than the alternative spot prices for soybeans. Our model is based on Zhang and Turvey (2019), where they use Vector Autoregressive Model to discover discontinuity pattern after the imposition of import tariff on U.S. soybeans. In this paper, I am going to enrich their analysis with data after the Phase One Deal in 2020, and I will include Brazilian soybean futures prices in the analysis. Most of the authors above agree that the CME dominates world markets in terms of price signalling. When there are price changes in the U.S., then they are transmitted to Chinese and Brazilian markets. This has been true mostly for soybeans, which were not as heavily regulated in the past. However, some recent papers are coming to a new conclusion, showing there is a bidirectional signalling coming from China and Brazil. 10 CHAPTER 3 BACKGROUND The roots of the U.S.-China trade war can be traced back to the early 2000s, when China began to emerge as a major economic power. Over the past two decades, China has become the world’s largest exporter, accounting for approx- imately 14 percent of global exports in 2020.1 However, many U.S. policymakers have accused China of engaging in unfair trade practices, including intellectual property theft, forced technology transfers,2 and currency manipulation.3 In response to these concerns, the U.S. government has taken several mea- sures aimed at curbing China’s economic influence. In 2018, the Trump adminis- tration, imposed tariffs on $50 billion worth of Chinese goods, citing intellectual property theft and other issues.4 The following year, the U.S. increased tariffs on an additional $200 billion worth of Chinese goods, leading China to retaliate with tariffs of its own.5 Since the initial imposition of tariffs, the U.S. and China have engaged in a series of retaliatory measures. In addition to tariffs, the U.S. has blacklisted Chi- 1Nicita, A., & Razo, C. (2021, April 27). China: The rise of a trade titan. UNCTAD. Retrieved from https://unctad.org/news/china-rise-trade-titan 2Carnegie Endowment for International Peace. (2022, April 25). Countering Unfair Chi- nese Economic Practices and Intellectual Property Theft [Policy Outlook]. Retrieved from https://carnegieendowment.org/2022/04/25/countering-unfair-chinese-economic-practices- and-intellectual-property-theft-pub-86925 3U.S. Department of the Treasury. (2019, August 5). Treasury Designates China as a Cur- rency Manipulator [Press release]. Retrieved from https://home.treasury.gov/news/press- releases/sm751 4Wei, Z.,& Lu, Y. (2018, July 13). China’s Trade Surplus With U.S. Hits Record as Fight In- tensifies. Bloomberg. https://www.bloomberg.com/news/articles/2018-07-13/china-s-trade- surplus-with-u-s-hits-record-as-fight-intensifies 5Davis, B., & Smithers, R. (2019, May 6). U.S.-China trade war: Trump raises tariffs on $200bn of Chinese goods – business live. The Guardian. https://www.theguardian.com/business/live/2019/may/06/china-us-trade-war-tariffs- markets-economy-business-live 11 nese tech firms such as Huawei and TikTok, citing national security concerns.6 China has responded with its own blacklist of U.S. companies, including Apple and Boeing.7 The U.S. has also taken steps to limit Chinese investment in the U.S., in- cluding tightening restrictions on foreign investment in sensitive industries.8 In addition, the U.S. has increased scrutiny of Chinese companies listed on U.S. stock exchanges, leading to the delisting of several Chinese firms.9 China has responded to U.S. measures with its own restrictions on U.S. goods and services. In addition to retaliatory tariffs, China has placed restric- tions on U.S. agricultural products and banned U.S. tech firms from doing busi- ness with Chinese government agencies.10 The U.S.-China trade war has had significant implications for the global economy. The disruption of global supply chains has led to increased costs for businesses and consumers, particularly in industries such as technology and agriculture.11 In addition, the trade war has led to a slowdown in global eco- nomic growth, with the International Monetary Fund (IMF) estimating that the conflict could reduce global GDP by up to 0.8 percent.12 6Luhby, T. (2020, January 15). The U.S.-China trade war isn’t over yet. CNN Business. https://www.cnn.com/2020/01/15/business/us-china-trade-war-phase-one-deal/index.html 7CNBC. (2021, March 18). U.S.-China Trade War. https://www.cnbc.com/us-china-trade- war/ 8Luhby, T. (2020, January 15). The U.S.-China trade war isn’t over yet. CNN Business. https://www.cnn.com/2020/01/15/business/us-china-trade-war-phase-one-deal/index.html 9CNBC. (2021, March 18). U.S.-China Trade War. https://www.cnbc.com/us-china-trade- war/ 10CNBC. (2021, March 18). U.S.-China Trade War. https://www.cnbc.com/us-china-trade- war/ 11OECD. (2019). Economic Outlook for Southeast Asia, China and India 2020: Towards Smart Urban Transportation. OECD Publishing. https://doi.org/10.1787/9789264312317-en 12IMF. (2021). World Economic Outlook, April 2021: Managing Divergent Recoveries. IMF Publishing. https://www.imf.org/en/Publications/WEO/Issues/2021/03/23/world- economic-outlook-april-2021-managing-divergent-recoveries 12 The trade war has also had geopolitical implications, with the U.S. and China engaging in a broader struggle for influence in the Asia-Pacific region. The U.S. has sought to strengthen ties with regional allies such as Japan and South Ko- rea, while China has sought to expand its influence through its Belt and Road Initiative (BRI) and other initiatives.13 13Schott, J. J. (2021, January 14). Trump’s China Tariffs: Successes, Failures, and Lessons. Council on Foreign Relations. https://www.cfr.org/blog/trumps-china-tariffs-successes- failures-and-lessons 13 CHAPTER 4 COUNTRY SPECIFIC SOYBEANS PRODUCTION The U.S.-China soybean trade war began in July 2018 when the U.S. imposed a 25 percent tariff on $34 billion worth of Chinese goods.1 In response, China implemented a 25 percent tariff on U.S. soybeans, one of its key agricultural imports from the U.S.. The move was part of a broader trade dispute between the two countries that had been brewing for months. Soybeans are an important commodity for both the U.S. and China. The U.S. is the world’s largest producer and exporter of soybeans, while China is the world’s largest importer of soybeans.2 In 2017, the U.S. exported $21.6 bil- lion worth of soybeans to China, accounting for more than half of its soybean exports.3 However, with the imposition of the tariffs, the trade flow was dis- rupted, and U.S. export prices decreased by about $0.74 per bushel according to Aaron Smith. But the U.S. farmers did not suffer the consequences of the tariffs due to U.S. the Federal government aid for farmers in a form of $1.65 per bushel in 2018, which is more than double the estimated damage. He also argues that farmers received several times more than what they lost.4 The tariffs had a significant impact on U.S. soybean farmers, who saw the price of their crops drop by almost 20 percent in the wake of the trade war. This was due to China’s decision to shift its soybean imports away from the U.S. 1Morrison, W. M., & Weiss, M. A. (2020). The U.S.-China Trade War: A Timeline. Congres- sional Research Service. https://fas.org/sgp/crs/row/R45621.pdf 2USDA. (2021). Foreign Agricultural Service. https://www.fas.usda.gov/topics/trade- data-analysis/exports-commodities-country 3Kuo, L., & Holmes, K. (2018). U.S.-China trade war: Soybean farmers are hurting, and they blame Trump. CNN Business. https://www.cnn.com/2018/07/19/economy/trump-china- trade-war-soybeans-farmers/index.html 4Smith, A. (n.d.). Trade War [Webpage]. Retrieved from https://asmith.ucdavis.edu/news/trade-war 14 to other countries, such as Brazil and Argentina, which are also major soybean exporters. This increased Brazilian soybean prices by about $0.97 per bushel. This shift in demand had a ripple effect throughout the U.S. soybean industry, as farmers struggled to find alternative markets for their crops. The soybean tariffs also had a broader impact on the U.S. economy. The U.S. agricultural sector, which accounts for a significant portion of the country’s ex- ports, and the prices of other agricultural commodities, such as corn and wheat, also fell as a result.5 Additionally, the soybean tariffs contributed to an over- all slowdown in trade between the two countries, as other industries were also affected by the ongoing dispute. The U.S.-China soybean trade war has had significant consequences for both countries, with the U.S. soybean industry bearing the brunt of the impact. While there have been some attempts to reach a resolution to the ongoing trade dis- pute, such as the signing of the phase one trade deal in 2020, the tariffs on soy- beans and other goods remain in place, leaving U.S. and Chinese economies to continue feeling the effects of the trade war.6 4.1 U.S. Soybeans When commodities across different countries are discussed, it is often incor- rectly assumed they are the same. Also, in the case of soybeans, there are differ- 5Srinivasan, A. (2021). The Impact of the U.S.-China Trade War on Agricultural Commodi- ties. Trade Data Monitor. https://tradedatamonitor.com/the-impact-of-the-us-china-trade- war-on-agricultural-commodities/ 6Peterson Institute for International Economics. (n.d.). U.S.-China Phase One Tracker: China’s Purchases of U.S. Goods [Webpage]. Retrieved from https://www.piie.com/research/piie-charts/us-china-phase-one-tracker-chinas-purchases- us-goods 15 ences in U.S. and Brazilian soybeans that may add to friction, when it comes to price signalling across different markets. U.S. soybeans are known to be mostly genetically modified (GMO) crops according to U.S. Department of Agriculture (USDA).7 Their protein content is around 36 percent8 and growing season is from spring to autumn. Soybeans harvest starts in September and ends in November. From the diagram be- low, you can see the states and counties, in which soybeans are predominantly grown. The states with the largest soybean production are Iowa, Illinois, and Min- nesota. According to USDA, just over 70 percent of the soybeans grown in the United States are used for animal feed, with poultry being the number one livestock sector con- suming soybeans, followed by hogs, dairy, beef and aquaculture.11 The second largest market for U.S. soybeans is for human consumption for saladas and oils, which consists of 15 percent of the U.S. soybean crop. Then about 5 percent is used for biodiesel. 7U.S. Department of Agriculture Economic Research Service. (2021). Adoption of Genetically Engineered Crops in the U.S.: Recent Trends in GE Adoption. Retrieved from https://www.ers.usda.gov/data-products/adoption-of-genetically-engineered-crops-in- the-u-s/recent-trends-in-ge-adoption/ 8US Department of Agriculture, FoodData Central. (n.d.). Food Details - Soybeans, ma- ture seeds, raw. USDA FoodData Central. https://fdc.nal.usda.gov/fdc-app.html#/food- details/174270/nutrients 10United States Department of Agriculture, National Agricultural Statistics Service. (n.d.). Soybeans, Planted and Harvested, 2020. Retrieved from https://www.nass.usda.gov/Charts and Maps/graphics/SB-PR-RGBChor.pdf 11U.S. Department of Agriculture. (2015, February). Coexistence Fact Sheets: Soybeans. Retrieved from https://www.usda.gov/sites/default/files/documents/coexistence-soybeans- factsheet.pdf. 16 Figure 4.1: Soybeans 2022 Production by County10 4.2 Brazilian Soybeans Only about 2 percent of planted Brazilian soybeans are non-GMO, and with a current trend to increase their non-GMO soybeans acreage to satisfy increasing demand for non-GMO crops from Europe.12 Their protein content varies from year to year. While some say that Brazilian soybeans have on average higher 12Samora, R. (2022, June 23). Brazil to boost non-GMO soy area for higher European demand, industry group says. Reuters. Retrieved from https://www.reuters.com/article/brazil-gmo- soy/brazil-to-boost-non-gmo-soy-area-for-higher-european-demand-industry-group-says- idUKL1N2YA2P3 17 protein content than U.S. soybeans,13 some say they are about the same.14 But it is believed that Brazilian soybeans have a higher probability of having more protein content that the U.S. counterpart due to its vast Amazon landscape that can be used for crops. Brazilian soybeans are usually planted during the rainy season from October to December and their harvest typically starts in March and ends in June.15 As you can see in the Figure 4.2, soybeans are grown mostly inland and far from the coast. Transportation adds cost to Brazilian soybeans, and it happened that Brazilian soybeans production cost was lower than in the U.S. in 2006 and 2010, but due to transportation cost, the landed U.S. soybeans in China cost less than those from Brazil, as can be seen in figure 4.4.18 In Figure 4.3, we see that the development of BR-163 led to better intercon- nectedness of Mato Grosso, the new primary soybean production land with the ports in Santarem and Santos. After BR-163 pavement completion in 2017, ma- jority of soybeans were transported to the northern ports instead, which halved 13Stratfor. (2018, April 10). Why China Is Hungry For Brazilian Soy. Forbes. Retrieved from https://www.forbes.com/sites/stratfor/2018/04/10/why-china-is-hungry-for-brazilian- soy/?sh=49f6f785321d 14Grieshop, C. M., & Fahey, G. C. Jr. (2001). Comparison of Quality Characteristics of Soy- beans from Brazil, China, and the United States. Journal of Agricultural and Food Chemistry, 49(6), 2669-2673. 15Southern Ag Today. (2022, December 19). What to Expect from Brazil’s Soybean Crop. Re- trieved from https://southernagtoday.org/2022/12/19/what-to-expect-from-brazils-soybean- crop/ 17Figure 2 taken from: United States Department of Agriculture. (n.d.). Brazil - Soybean. Retrieved from https://ipad.fas.usda.gov/countrysummary/Default.aspx?id=BR&crop=Soybean 18Irwin, S., & Good, D. (2012). High-Frequency Trading and Soybean Futures Prices: A Case of Agricultural Malfunction? CARD Ag Policy Review, 2012(2), 1-14. Retrieved from https://www.card.iastate.edu/ag policy review/article/?a=132 20Irwin, S., & Good, D. (2012). High-Frequency Trading and Soybean Futures Prices: A Case of Agricultural Malfunction? CARD Ag Policy Review, 2012(2), 1-14. Retrieved from https://www.card.iastate.edu/ag policy review/article/?a=132. 22Irwin, S., & Good, D. (2012). High-Frequency Trading and Soybean Futures Prices: A Case of Agricultural Malfunction? CARD Ag Policy Review, 2012(2), 1-14. Retrieved from https://www.card.iastate.edu/ag policy review/article/?a=132. 18 Figure 4.2: Brazilian Soybeans Production17 trucking time from 3 days to 1.5 day. Better transport infrastructure from Mato Grosso to Santarem Port made landed Brazilian soybeans cheaper in China than the U.S. ones in 2020. 19 Figure 4.3: Brazil’s major road network in 2017 and BR-16320 Figure 4.4: Brazil’s Transportation Costs and Soybean Prices22 20 CHAPTER 5 INTERNATIONAL TRADE OF SOYBEANS & CHINA Figure 5.1 demonstrates that the global soybean trade is mainly dominated by the U.S. and Brazil. The cyclical nature of soybean imports from these countries is due to their opposite hemisphere locations. This causes different harvest- ing seasons. While U.S. harvests its soybeans between September and October1 while Brazil does so between March and June.2 This causes the U.S. to export soybeans from September to February, while Brazil’s export season runs from March to August.3 On the other hand, the Rest of the World (ROW) has a minor role in soybean trade with China. As a result, China has maintained a constant supply of soybeans throughout the year by importing them from both regions. China is known to consume about 110 million metric tons (MMT) of soy- beans per year.4 China is the world’s largest importer of soybeans, with an annual import volume of approximately 95 MMT5 accounting for about two- thirds of global soybean trade. China’s soybean consumption has been grow- ing rapidly due to the country’s expanding hog industry. Animal feed and oil accounts for approximately 85 percent of China’s soybean imports, with only 1USDA National Agricultural Statistics Service. (2010). Usual Plant- ing and Harvesting Dates for U.S. Field Crops. Retrieved from https://usda.library.cornell.edu/concern/publications/vm40xr56k 2USDA Foreign Agricultural Service. (n.d.). Brazil Crop Calendar. Retrieved from https://ipad.fas.usda.gov/rssiws/al/crop calendar/br.aspx 3CME Group. (2020). The Relationship Between Global Soybean Prices. Retrieved from https://www.cmegroup.com/education/articles-and-reports/relationship-between-global- soybean-prices.html 4South China Morning Post. (2021, December 3). China’s soybean output tops 20 million tonnes for the first time as it shifts away from import reliance. Retrieved from https://www.scmp.com/economy/china-economy/article/3207577/chinas- soybean-output-tops-20-million-tonnes-first-time-it-shifts-away-import-reliance 5South China Morning Post. (2021, November 30). China’s soybean imports have peaked and will keep falling, Beijing stresses food security, report says. Retrieved from https://www.scmp.com/economy/china-economy/article/3206301/chinas-soybean-imports- have-peaked-and-will-keep-falling-beijing-stresses-food-security-report-says 21 Figure 5.1: Chinese Monthly Soybean Imports about 15 percent used for human consumption.6 The remaining soybean im- ports are used for various purposes such as vegetable oil production and direct human consumption. In Figure 5.2, you can see the comparison of Chinese monthly consumption to total soybean production and imports. The red curve shows a steady increas- ing soybean consumption. However, there have been a few unforeseen factors that have reduced the demand for soybeans in China, one of which is the out- break of African swine fever (ASF). The first instance of ASF was confirmed in 6South China Morning Post. (2021, September 12). China’s soybean output tops 20 million tonnes for the first time as it shifts away from import reliance. South China Morning Post. https://www.scmp.com/economy/china-economy/article/3207577/chinas- soybean-output-tops-20-million-tonnes-first-time-it-shifts-away-import-reliance 22 China during August of 2018,7 which coincided with the initial round of tariffs that were imposed on U.S. soybean exports. While it has been difficult to esti- mate the extent of losses to China’s hog herd due to ASF, Iowa State University researchers believe that there has been a 14 percent decline in pork production in China as a direct result of this disease.8 You can see in Figure 5.3, by September 10, 2019, there were 143 outbreak recorded.9 As a result, it is expected that the quantity of soybeans needed for hog feed would decrease by 8 million metric tons.10 Another factor that contributed to this decrease in demand for soybean im- ports in China is the updated feed standards. In October of 2018, the China Feed Industry Association introduced new standards for swine and poultry feed, which lowered crude protein levels by 1.5 and 1 percent respectively.11 The Chinese Ministry of Agriculture stated that these new standards could poten- tially reduce China’s annual soybean consumption by 14 million metric tons.12 Thus, the combination of ASF and lower protein requirements for animal feed could have resulted in a reduction of 22 million metric tons in Chinese demand for soybeans. 7Shao, B., Li, Y., Wang, H., Han, S., & Li, X. (2018). African swine fever in China: An overview. Transboundary and Emerging Diseases, 65(5), 1073-1076. 8Zhang, W., Mapemba, L. D., & Zheng, Z. (2019). The Impact of African Swine Fever on China’s Soybean and Corn Imports. International Journal of Agricultural Economics, 4(3), 83- 95. 9Reuters. (2019). China’s pig farmers struggle to rebuild herds in wake of African swine fever. Retrieved from https://www.reuters.com/graphics/CHINA-SWINEFEVER- FARMERS/010090DR0KM/index.html 10Cowley, C., & Arita, S. (2019). Reshuffling in Soybean Markets following Chinese Tariffs. Amber Waves: The Economics of Food, Farming, Natural Resources, and Rural America, 17(2), 1-8. 11Sun, X., Pan, J., & Chiang, L. C. (2018). China’s Soybean Future Outlook: A Perspective from Feed Industry. Global Agricultural Information Network Report, GAIN Report Number CH18053. 12Ministry of Agriculture of the People’s Republic of China. (2018). Announcement on the Revision of the National Feed Standard for Swine and Poultry. Beijing: Ministry of Agriculture of the People’s Republic of China. 14Reuters. (2019). China’s pig farmers struggle to rebuild herds in wake of African 23 Figure 5.2: Chinese Monthly Soybeans Consumption, Production and Im- ports Figure 5.3: ASF Outbreaks in China14 24 The decline in U.S. exports to China serves as evidence for the reduced de- mand. U.S. soybean exports to China in 2018 dropped to 8.2 million metric tons, which is approximately 22 million metric tons less than the previous four-year average. In the meantime, China’s soybean imports from the rest of the world, particularly Brazil, were increasing.15 Therefore, the unexpected factors such as the outbreak of ASF and updated feed standards have led to a significant reduction in Chinese demand for soy- beans. The decline in U.S. exports to China supports this notion, as Brazil’s soybean imports to China continue to grow. The high demand for soybeans as animal feed in China is primarily driven by the country’s booming hog industry, which has been expanding rapidly over the years. The industry’s demand for soybeans is expected to continue to grow in the coming years, driven by factors such as population growth, increasing de- mand for meat, and changes in dietary habits. As a result, China’s dependence on soybean imports is likely to remain high in the foreseeable future. swine fever. Retrieved from https://www.reuters.com/graphics/CHINA-SWINEFEVER- FARMERS/010090DR0KM/index.html 15Cowley, C., & Arita, S. (2019). Reshuffling in Soybean Markets following Chinese Tariffs. Amber Waves: The Economics of Food, Farming, Natural Resources, and Rural America, 17(2), 1-8. 25 CHAPTER 6 THE ECONOMIC FRAMEWORK 6.1 Trade Theory and Comparative Advantage Adam Smith in the The Wealth of Nations (1776) argued that each country should specialize in producing those commodities in which it has absolute advantage. Absolute advantage occurs when a country can produce at lower real cost than another country. David Ricardo, in his Principles of Political Economy and Taxa- tion (1817) argues that countries don’t need absolute advantage to benefit from trade. He argued that it is enough when a country has comparative advantage, which occurs when country can produce at lower opportunity cost. In Figure 6.1, you can see that the trade balnce with the world in blue, China in red, and Rest of the World (ROW) in green. We can see that the trade bal- ance with China and ROW seems to be highly correlated and moving together in the same direction. However, in the years from 2018 to 2020, we see that they diverged. The trade balance with China improved while it worsened with the ROW. In 2018, the trade war between the U.S. and China errupted and the U.S. stopped sourcing supplies from efficient and more competitive Chinese pro- ducers, but turned to producers in the ROW. Therefore, the trade balance with China improved, however, it overall deteriorated. This shows that trade balance cannot be solved through tariffs, if governments decide to pursue such policies. Another way to show what a country is efficient in producing is by calcu- lating its revealed comparative advantage (RCA). This is based on Ricardian 2Data taken from U.S. Census Bureau 26 Figure 6.1: U.S. Trade Balance from 1989 to 20222 Trade Theory. In essence, it is challenging to calculate compartive advantage, as it is very complex to account for every aspect of production. However, revealed comparative advantage simplifies it and uses trade data to reveal such efficien- cies. The following formula provides a general indication of country’s export strength: RCA = US Soybean Exports US Total Exports World Soybean Exports World Total Exports = US Soybean Export Share Soybeany Export Share with World Trade It is important to note however, that national measures which affect competi- tiveness such as tariffs, non-tariff measures, subsidies and others are not taken into account in the RCA metric. Using the RCA formula we arrive at the following results for year 2017 for U.S. and Brazilian soybeans producers: U.S. at 4.11 and Brazil at 46.37. Value 27 greater than 1 means that the country has RCA in production of that commodity. Brazil seems to have higher RCA than the U.S., which can be explained through the Heckscher-Ohlin Model (H-O Model). According to H-O Model, country will specialize in production of that product, in which it has abundant produc- tion resources. Now the model specifies capital and labour in particular. Brazil is known to have vast land (Mato Grosso) and labour resources. Additionally, keeping in mind the idea of comparative advantage and opportunity cost, the U.S. can more effectively use its scarce resources to produce more technologi- cally advanced goods, which also contributes to its lower RCA. 6.2 Tariff and World Prices The economic signaling between a dominant demand center and a dominant supply center relies on informational signaling across market boundaries to es- tablish full transparency in price discovery. We set up the economic framework using classical three-panel trade diagram where an import tariff reduces world prices to PX for the exporter to and increase domestic prices in China to PM (trade levels decline). Figure 6.2 implies the price transmission between U.S. and Chinese market prices are now less than 1 due to the tariff, the extent to which depends on the relative elasticities of excess supply and demand, and on the level of the tariff. This has important implications for price discovery in our statistical analysis to follow. Furthermore, it was not immediately obvious in June 2018 that in addition to the tariffs, China would retaliate further with non-tariff measures, the most 28 Figure 6.2: The effect of an ad valorem import tariff on U.S. and China’s Soybean Market significant being embargo and the moral suasion of Chinese crushers to increase imports from Brazil at the expense of U.S. orders (this being reflected in the price premium for Brazilian soybeans over U.S. prices for the six-month time period ending in November 2018). The effect of an import embargo on U.S. soybean exports would sever the link between U.S. and Chinese market prices, just like an import quota would in the supply/demand analysis in Figure 6.2: the excess demand curve facing the U.S. would be vertical. This reduces U.S. exports even further, putting greater downside pressure on U.S. soybean prices This is a direct consequence of commitments by (state and private) importers, crushers, and further processors to refuse U.S. soybeans regardless of the price. Consequently, the total trade volume decreased, together with a higher import price in China and a lower export price in the U.S. Consequently, price signals between Chinese and U.S. soybeans weaken even further, and the U.S. domestic soybean price would no longer effectively influence international and Chinese markets. This is core to the economic analyses presented in this paper. 29 Figure 6.3: Pre-import tariff, Brazil’s export are large enough to fill the void if China reduces demand for U.S. soybeans Figure 6.2 assumes that China has no additional source of soybeans to im- port. However, as shown in Figure 6.3 China’s demand exceeded that exported by the U.S. with the bulk of the supply shortage coming from Brazil. Muham- mad et al (2018; Table 3), for example, argue that supply shortages due to tar- iffs on U.S. soybeans would be replaced on almost a 1:1 basis by Brazil in the short run. This is precisely what happened. To show the impacts of a compet- ing exporting fulfilling the void, consider Figure 8. Brazil’s exports are large enough to backfill China’s reduced imports from the U.S. if an import tariff is imposed (not shown). So why was there a price premium on Brazilian soybean prices for six months ending in November 2018? It was likely due to the need to fulfill China’s soybean processors demand for soybeans until the next U.S. soybean harvest (and in anticipation of the Southern hemisphere production months later). 30 6.3 Empirical Evidence on U.S. and Brazilian Spot Prices When examining the spot prices for U.S. and Brazilian soybeans, according to the economic theory, we expect that U.S. soybeans prices will fall while Brazil- ian prices will increase due to increased demand. Figure 6.4 demonstrates how closely both spot prices follow each other until around June 2018, a month be- fore the import tariffs take effect. The vertical black line indicates the July 6 2018 date when the tariffs took effect. Then we see a divergence between the two prices due to the Chinese embargo on U.S. soybeans. The embargo was alter lossened in early 2019, due to Brazilian inability to fully satisfy Chinese demand for bad weather. Figure 6.4: U.S. and Chinese Spot Prices As we will see, the import tariffs had a ripple effect across storate and finan- 31 cial markets, which we will examine next. 32 CHAPTER 7 TRADE WAR AND GRAINS STORAGE The U.S.-China soybean trade war had a significant impact on U.S. grain stor- age. Due to the tariffs imposed by China on U.S. soybean imports, the demand for U.S. soybeans fell sharply, leading to a glut of soybeans in the U.S. market. This situation caused significant storage challenges for the U.S. grain industry, which had to find alternative storage options to handle the excess soybean sup- ply. According to a report by the USDA, the U.S. had a record soybean inventory of 3.74 billion bushels in the 2018/19 marketing year due to the trade war, up from 2.11 billion bushels in the previous year.1 The excess supply of soybeans resulted in storage challenges for the U.S. grain industry. Many grain elevators and storage facilities were at or near capacity, and some were forced to turn away soybean deliveries. The glut of soybeans also caused storage prices to rise, further impacting U.S. grain storage. As the supply of soybeans exceeded storage capacity, storage prices increased, making it more expensive for farmers and grain handlers to store their soybeans. This situation led to a shift in the way farmers stored their crops, with many opting to store their soybeans in bags or on-farm storage 1United States Department of Agriculture (USDA). (2019, October 10). Grain: World Markets and Trade. Foreign Agricultural Service. Retrieved from https://apps.fas.usda.gov/psdonline/circulars/grain.pdf Nelson, Heather A., and Ja- son Grant. “Understanding the U.S.-China Trade War: What Are the Stakes for Agri- culture?” Choices 34, no. 4 (2019): 1-6. https://www.choicesmagazine.org/choices- magazine/theme-articles/the-us-china-trade-war/understanding-the-us-china-trade-war- what-are-the-stakes-for-agriculture. Larson, Donald F., and Wendong Zhang. “The U.S.-China Trade War and Soybeans: The Harder the Fall, the Higher the Bounce?” Center for Agricultural and Rural Development (CARD) Policy Brief 19-PB 11 (2019). https://www.card.iastate.edu/products/publications/pdf/19pb11.pdf. 33 rather than in traditional grain storage facilities. The trade war also impacted the transportation of soybeans, adding to stor- age challenges. Due to the tariffs, the transportation of soybeans from the U.S. to China was less frequent, leading to a backlog of shipments waiting to be ex- ported. This situation added to the storage challenges faced by the U.S. grain industry, as it became more difficult to move excess soybeans from storage fa- cilities to ports for export. The storage challenges faced by the U.S. grain industry had a knock-on effect on U.S. futures markets. As the supply of soybeans exceeded storage capacity, storage prices increased, leading to a reduction in the price of soybean futures contracts. The oversupply of soybeans also impacted the basis, the difference between the local cash price and the futures price, which widened due to the excess supply of soybeans. The impact of the U.S.-China soybean trade war on U.S. grain storage was significant. The glut of soybeans resulting from the trade war led to storage challenges for the U.S. grain industry, with many storage facilities at or near capacity. The situation also impacted transportation, making it more difficult to move excess soybeans from storage facilities to ports for export. The storage challenges faced by the U.S. grain industry had a knock-on effect on U.S. futures markets, leading to a reduction in the price of soybean futures contracts and a widening of the basis.2 Basis = Cash Price - Futures Price where: 2Irwin, S. H., & Good, D. L. (2013). Solving the Market’s Non-convergence Puzzle. farmdoc daily, 3(151). Retrieved from https://farmdocdaily.illinois.edu/2013/08/solving-markets-non- convergence-puzzle.html 34 Cash Price: the current market price of a commodity in a specific location. Futures Price: the price of the same commodity in the futures market for a specific delivery month. The basis is a measure of the deviation of the cash price from the futures price. It reflects the local supply and demand conditions for the commodity, as well as transportation costs and other factors that affect the price in a particular location. 7.1 Soybean Futures Nonconvergence Figure 7.1: U.S. Soybeans Nonconvergence Figure 7.1 shows the difference between the cash price of soybeans and future contracts expiring in 2018 and 2019, also known as basis. 35 Futures prices tend to converge towards the spot prices when approaching its last trading day. According to theory, spot price should be the same as fu- tures prices at the expiry, as we see it for the 2019 contract. However, 2018 fu- tures contract shows us a different scenario because we see that soybeans fail to converge. Futures prices remain higher than spot prices, even though the future contract is eligible for delivery, same as spot contracts. According to USDA and researchers from the University of Illinois, the main reason for non-convergence is the storage fee.3 At the expiry of the futures contract, you not necessarily re- ceive the grains to your doorstep, but you receive a delivery instrument. Those holding delivery instruments would act on their best self-interest not receiving the grains and using the futures market to subsidise their storage costs. How exactly? Firstly, as the holder of the delivery instrument, you don’t have a spe- cific date by which to get the grains delivered. Secondly, when there is a market shock and storage costs become excessively high, it is better not to store those physical grains yourself, but to use the futures market and pay a storage fee. The sudden shock in 2018 with Chinese tariffs caused a spike in demand for storage for millions of tons of soybeans. 3Irwin, S. H., & Good, D. L. (2013). Solving the Market’s Non-convergence Puzzle. farmdoc daily, 3(151). Retrieved from https://farmdocdaily.illinois.edu/2013/08/solving-markets-non- convergence-puzzle.html 36 7.2 Corn Futures Nonconvergence Figure 7.2: Source: Bloomberg We know that soybeans and corn are highly substitutable and easily replace- able amongst each other in farming. The diagram above shows that also corn experienced nonconvergence under the strain of high demand for storage, as the storage space can be utilised by any grain. Therefore, the impact on conver- gence in futures market was widespread. But also in this case, the corn futures converged in 2019. 7.3 Economics of Nonconvergence Here is the economics behind the storage prices, inventories, and non- convergence. From the diagram below we can see the market for storage and 3Adjemian, M. K., Garcia, P., Irwin, S., & Smith, A. (2018). Non-Convergence in Domestic Commodity Futures Markets: Causes, Consequences, and Remedies. Applied Economic Per- 37 then the second market for cash price of grain and futures prices in period one and two. Figure 7.3: Convergence under Normal Inventory Demand5 In Figure 7.3, besides the equilibrium between storage and inventory, we also see the PI which stands for physical storage cost and Pbar stands for delivery instrument storage fee. In this case, we see that delivery instrument storage fee is higher than the one for physical storage, hence the market converges, as we can see in the second panel. In the second panel, the difference between the spot price at t0 and those at t1 and t2 is PI , which is the physical storage cost. This is how futures converge under normal market conditions. In our case with soybeans in 2018, the market experienced a significant demand for storage that shifted the demand curve and created a new equilibrium PII above Pbar. PII spectives and Policy, 40(1), 31-53. doi: 10.1093/aepp/ppx049 5United States Department of Agriculture (USDA), Economic Research Ser- vice. (2013). Non-Convergence in Domestic Commodity Futures Markets: Causes, Consequences (Economic Information Bulletin Number 115). Retrieved from https://www.ers.usda.gov/webdocs/publications/43777/39376 eib115.pdf 38 means that storage and inventory hits equilibrium in the second period, not in the year when the market is supposed to clear. Figure 7.4: Nonconvergence under Excess Inventory Demand7 In this case, hedgers, investors, and speculators would prefer to hold on to delivery instruments and take on storage fees. These extra storage fees that usu- ally don’t occur under normal market circumstances, create a wedge between the future price and the spot price in t1. Essentially, the price of the future con- tract becomes decoupled from the market by this wedge (w). Recalling the soy- beans cash – futures for 2018 and 2019 depicts this specific scenario above. The initial demand for storage was overwhelming for the market, which could not be accommodated and resulted in this nonconvergence, which is the wedge or delivery instrument storage fee. But in 2019, the futures market for soybeans converged. 7United States Department of Agriculture (USDA), Economic Research Ser- vice. (2013). Non-Convergence in Domestic Commodity Futures Markets: Causes, Consequences (Economic Information Bulletin Number 115). Retrieved from https://www.ers.usda.gov/webdocs/publications/43777/39376 eib115.pdf 39 CHAPTER 8 DATA 8.1 Price Discovery The aim of this paper is to investigate how the imposition of import tariffs by China on U.S. exports of soybeans affected the flow of information and mar- ket signaling between the U.S., Brazil, and China. While previous studies have focused on settlement prices or closing prices, our analysis seeks to compre- hensively understand the patterns and prices in these markets in order to fully uncover the relationship between the futures markets. The Chinese futures mar- ket operates from 9:00 am to 11:30 am and from 1:30 pm to 3:00 pm, while the Chicago market opens at 8:30 a.m. and closes at 1:20 p.m., with overnight GLOBEX trading between 7:00 p.m. and 7:45 p.m. Central Time. We use the CME final or settled price reported around 3pm daily to compare and analyze the markets. This method is the same as the one used in Zhange and Turvey (2019). There is a time difference of 7.5 hours between the Chicago close and DCE open and a 5.5-hour difference between the DCE close and the CME open. Therefore, the opening bid in China, prior to the import tariff, would receive as an informational signal the closing or settle price in Chicago 7.5 hours pre- vious. Similarly, the opening bids in Chicago would rely on the information content contained in the closing prices in Dalian, 5.5 hours previous. For Brazil- ian soybeans, we use Cash-Settled Soybean Futures Contract at the Price of the CME Group Mini-Sized Soybean Futures Contract (SJC). This market opens at 6:00 a.m. and closes at 1:20 p.m. Central Time. Thus, the Brazilian soybean 40 market opens 2.5 hours ahead of the Chicago market, and this can send price signals to the Chicago opening market. As for the Chinese market, Brazil is 12 hours behind Dalian time. Therefore, we can test B3 closing prices against next day’s DCE opening prices, and DCE closing prices with B3 opening prices. Due to the time difference between the two countries, the Chinese futures market, Brazilian, and the U.S. futures mar- ket do not trade simultaneously. The trades for soybean futures on the CME are typically executed from 8:30 a.m. to 1:20 p.m. Central Time, while in Brazil, it starts at 6:00 a.m. and closes at 1:20 p.m. Central Time. This time period is not a trading window for China. The DCE, on the other hand, is open from 9:00 a.m. to 3:00 p.m., during which period the exchange accepts and matches orders for soybean futures. As the two markets trade in a sequential order, traders will look at the performance in the other country and make their own trading strate- gies for the day, thereby integrating prices and markets. Thus, the relationship between the markets is a sequential one. In this paper, we will address this issue by co-integrating different price pairs according to sequence of time: 1. CME closing price and DCE opening price 2. DCE closing price and CME opening price 3. B3 closing price and DCE opening price 4. DCE closing price and B3 opening price 41 8.2 Descriptive Statistics We collected data on soybean futures and spot prices for the U.S., Brazilian, and Chinese markets from the Bloomberg terminal. The data includes opening price, closing price, and settlement price of the CME, DCE (soybean #1 futures contract), and B3 Cash-Settled CME Mini-Sized Soybean Futures. We retrieved date-matched time series data from January 4, 2016, to April 20, 2023, cover- ing periods before and after the imposition of import tariffs on U.S. exports. However, due to differences in holiday periods, some data did not correspond to trading dates in the U.S. or China. To address this issue, we used linear interpolation to fill in single-day missing data and deleted unmatched daily ob- servations missing for more than two successive days. Table 1 shows the data observations. We analyzed three periods: the Pre-Trade War period from January 2016 to July 2018, the Import Tariff period from July 2018 to January 2020, and the Post- Trade War period from January 2020 to April 2023. Figure 8.1 provides a visual representation of these periods. Our aim is to examine how the imposition of import tariffs by China on U.S. exports of soybeans affected the information flows and market signaling between the three countries. 42 Data pairs Number of observations Period I: Pre-Trade War CME close & DCE open 474 DCE close & CME open 475 B3 close & DCE open 475 DCE close & B3 open 475 Period II: Trade War CME close & DCE open 315 DCE close & CME open 316 Period III: Post-Import Tariff (Jan 2020 onwards) CME close & DCE open 518 DCE close & CME open 516 Period II + Period III for Brazil B3 close to DCE open 1052 DCE close to B3 open 1051 Table 8.1: Data Description 43 Figure 8.1: Time Series and Trade War Periods cme open cme close dce open dce close brz open brz close Min. 2.067 2.068 2.515 2.515 2.085 2.082 1st Qu. 2.199 2.199 2.654 2.655 2.203 2.202 Median 2.303 2.303 2.782 2.782 2.306 2.305 Mean 2.394 2.394 2.88 2.88 2.393 2.393 3rd Qu. 2.629 2.632 3.151 3.152 2.623 2.619 Max. 2.874 2.873 3.313 3.31 2.872 2.873 sd 0.228 0.228 0.252 0.252 0.222 0.222 skewness 0.561 0.556 0.266 0.265 0.551 0.548 kurtosis 1.825 1.817 1.475 1.472 1.837 1.831 Unit Root Tests ADF 0.01 0.01 0.01 0.01 0.01 0.01 KPSS 0.1 0.1 0.1 0.1 0.1 0.1 Table 8.2: Summary statistics and unit root tests for First Order Differ- enced Data Series in log form. 44 CHAPTER 9 METHODOLOGY 9.1 RVAR Model Following the structure from Zhang and Turvey (2019), we use RVAR model, which is a time-series econometric model used to analyze the dynamic relation- ships among multiple variables over time. It is a popular model in macroeco- nomic analysis and forecasting, and is particularly useful in studying the inter- dependence and feedback effects between variables. y1,t = α1 + β1,1 · y1,t−1 + β1,2 · y2,t−1 + β1,3 · y3,t−1 + ε1,t y2,t = α2 + β2,1 · y1,t−1 + β2,2 · y2,t−1 + β2,3 · y3,t−1 + ε2,t y3,t = α3 + β3,1 · y1,t−1 + β3,2 · y2,t−1 + β3,3 · y3,t−1 + ε3,t The RVAR model is a system of simultaneous equations, where each vari- able is regressed on its own lagged values and the lagged values of all the other variables in the system. This allows the variables to influence each other and capture the interdependence among them. The equations in the system are es- timated together using a maximum likelihood method or Bayesian estimation, and the parameters of the model provide insights into the direction and strength of the relationships between the variables. One of the advantages of the RVAR model is that it is a data-driven approach that does not require prior assumptions about the underlying economic theory or the direction of causality between variables. Instead, it allows the data to speak for itself and identify the most significant relationships between the vari- ables. Additionally, the model can be easily expanded to include more variables 45 and to account for non-linearities and structural breaks in the data. The RVAR model makes several assumptions: 1. Stationarity: The time series data used in the RVAR model should be sta- tionary, meaning that the mean and variance of the series do not change over time. If the data is non-stationary, it needs to be transformed (e.g., differencing) to achieve stationarity. 2. Linearity: The relationships between the variables in the RVAR model are assumed to be linear. Nonlinear relationships would require a different modeling approach. 3. No perfect multicollinearity: The variables included in the RVAR model should not have perfect linear relationships with each other, as this would lead to multicollinearity issues. 4. No endogeneity: The variables in the model should not be correlated with the error terms. Endogeneity can lead to biased and inconsistent parame- ter estimates. 5. No serial correlation: The error terms in the RVAR model should not be serially correlated, meaning that there should be no pattern of residual autocorrelation. 6. Homoscedasticity: The error terms should have constant variance over time, known as homoscedasticity. Heteroscedasticity (varying variance) can lead to inefficient parameter estimates. 7. No misspecification: The RVAR model should be correctly specified, meaning that all relevant variables and lags are included, and no irrele- vant variables are included. 46 8. Residuals are normally distributed: The error terms should be normally distributed. It is worth noting that some of these assumptions can be relaxed or tested for violations, such as the stationarity assumption, which we tested using unit root tests. Choosing the correct number of lags in a Vector Autoregression (VAR) model is crucial for obtaining accurate and reliable results. The number of lags repre- sents the number of past observations that are included in the model as predic- tors for each variable. One commonly used approach for selecting the number of lags is the Akaike Information Criterion (AIC) or the Bayesian Information Criterion (BIC). These information criteria balance the goodness of fit of the model with the number of parameters used in the model. In general, the lower the AIC or BIC value, the better the model fits the data. Besides AIC, in this pa- per we also utilize the Hannan-Quinn criterion (HQ), Schwarz Criterion (SC), and the “Final Prediction Error” (FPE). 9.2 Granger Causality Tests The test was founded by Clive Granger, a British economist, in the 1960s. The idea behind the test is to find out if one time series is a good predictor of other time series. I used the test to find out the prediction power of our VAR model. Another aspect of Granger causality is that it tests prediction rather than eco- nomic causality itself. To prove causality in economic model, using Granger causality is not sufficient. In other words, if we say that some time series A was to have Granger-causal properties in relation to another time series B, it 47 means that time series A can predict time series B, even when all other relevant information is taken into account. To test for Granger causality, a commonly used approach involves estimating the autoregressive models for both variables separately, and then comparing the prediction error of each model. If the model including the lagged values of X has a lower prediction error than the model excluding those values, then X is said to Granger-cause Y. This approach assumes that causality should run from the cause to the effect, meaning that the cause (X) should precede the effect (Y) in time. Here are the results for our data series: Table 9.1: Granger Causality Test: Model 1 vs. Model 2 Model Res.Df Df F Pr(>F) dce open = Lags(dce open, 1:3) + Lags(cme close, 1:3) 1524 -3 3.6367 0.01242* cme open = Lags(cme open, 1:3) + Lags(dce close, 1:3) 1524 -3 3.9706 0.007844** dce open = Lags(dce open, 1:3) + Lags(brz close, 1:3) 1524 -3 4.2854 0.005075** brz open = Lags(brz open, 1:3) + Lags(dce close, 1:3) 1524 -3 3.4391 0.01629* Note: * indicates significance at the 0.05 level; ** indicates significance at the 0.01 level. From the above tests we see that all equations satisfy Granger-Causality test. It is important to note that Granger causality is a statistical concept and does not necessarily imply a true causal relationship in the sense of a physical or mechanistic connection. It also does not indicate the direction or mechanism of the causal relationship. Furthermore, it is possible for two variables to exhibit Granger causality in both directions, meaning that they are mutually causative. 48 CHAPTER 10 STATIONARITY 10.1 Autocorrelation Function (ACF) and Partial Autocorrela- tion Function (PACF) ACF and PACF models are used to identify the order of an autoregressive model. In our case, it also helps to visualize lag order and stationarity. However, it is different than ADF test, which specifically tests for stationarity. ACF and PACF assess the order of lags needed for VAR model to get the most accurate estimate. ACF measures the correlation between a variable and its lagged values. In other words, it shows persistence and the correlation between the variable and its past values at that lag. In our graphs, you are going to find very strong persistence, as the values for our time series data are well above 0.05. This can be seen for all three commodities exchange markets in the U.S., China, and Brazil. PACF, on the other hand, measures the correlation between a variable and its lagged values, after controlling for the effects of all shorter lags. PACF is used to identify the order of the AR model, where each lag in the PACF plot corresponds to the number of lags in the AR model. If there is a significant spike in the PACF plot at a specific lag, it suggests that there is a strong correlation between the variable and its past values at that lag, indicating that an AR term might be needed in the model. In our case, first order lag seems to be sufficient enough to eliminate the strong correlation between observations. At CME and B3 markets, we see some spikes that are after the first order lag. However, they 49 seem to be minimal, and this issue will be further dealt with in VAR model using other lag estimators such as Akaike Information Criterion (AIC). 2016 2017 2018 2019 2020 2021 2022 8 10 12 14 16 18 CME futures settlement price in USD/bushel Date C M E _C lo se CME Price 0.00 0.02 0.04 0.06 0. 0 0. 2 0. 4 0. 6 0. 8 1. 0 Lag A C F ACF for Last Price @ CME 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0. 0 0. 2 0. 4 0. 6 0. 8 1. 0 Lag P ar tia l A C F PACF for Last Price @ CME Figure 10.1: CME ACF and PACF 50 2016 2017 2018 2019 2020 2021 2022 15 20 25 DCE futures settlement price in USD/bushel Date D C E _C lo se DCE Price 0.00 0.02 0.04 0.06 0. 0 0. 2 0. 4 0. 6 0. 8 1. 0 Lag A C F ACF for Last Price @ DCE 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0. 0 0. 2 0. 4 0. 6 0. 8 1. 0 Lag P ar tia l A C F PACF for Last Price @ DCE Figure 10.2: DCE ACF and PACF 51 2016 2017 2018 2019 2020 2021 2022 8 10 12 14 16 Brz futures settlement price in USD/bushel Date B rz _C lo se BRZ Price 0.00 0.02 0.04 0.06 0.08 0. 0 0. 2 0. 4 0. 6 0. 8 1. 0 Lag A C F ACF for Last Price @ BRZ 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0. 0 0. 2 0. 4 0. 6 0. 8 1. 0 Lag P ar tia l A C F PACF for Last Price @ CME Figure 10.3: B3 ACF and PACF 10.2 Augmented Dickey-Fuller (ADF) Test With the Dickey-Fuller test, we can determine stationarity of time series data. It is a statistical test testing the unit root. It examines whether the mean, vari- ance remain constant over time. If that is the case, then we have a stationary time series data. Otherwise, we have non-stationary data series with trends and seasonalities that need to be further analysed with other tests. It helps with assessing the assumptions of the VAR model used in our analysis. 52 The Dickey-Fuller test is a hypothesis test that tests the null hypothesis that a time series is non-stationary against the alternative hypothesis that it is sta- tionary. The test involves estimating the following regression model: In the ADF, we test a null hypothesis that the time series is non-stationary against the alternative hypothesis that it is stationary. Therefore, we must reject the null hypothesis to prove stationarity. Below you is specified simple DF test: ∆Yt = α + βYt−1 + εt where Yt is the time series data, ∆Yt is the first difference of the data (i.e., the difference between each observation and the previous observation), Yt−1 is the lagged value of the data, α and β are constants, and εt is the error term. The ADF is mostly the same as the simple case, with the exception of adding further lags in the model to make the test more comprehensive, which is speci- fied below: ∆Yt = α + βYt−1 + βYt−2 + εt The intuition behind the Dickey-Fuller test is that if a time series is non- stationary, then it is likely to be sensitive to its own past values. In other words, if the current value of the time series depends on the previous value, then the time series is likely to be non-stationary. By taking the first difference of the data, the test removes any linear trend in the data, and by regressing the first difference on the lagged value of the data, the test checks whether there is any remaining dependence on the past values. 53 Figure 10.4: Time Series for Opening and Closing Soybean Prices for CME, DCE and B3 54 Figure 10.5: FOD Time Series The idea behind the test is that when we have present non-stationarity in our time series data, then our current observations are depended on the previous observations. In other words, past values predict future values for the same time series data. The way we deal with it, is by taking the difference with the past value(s). This in theory removes all the trends and dependencies on the past values present in our time series data set. The test statistic is then calculated based on the coefficient estimate for β and the standard error of the estimate. If the test statistic is less than the critical value 55 at a given significance level, the null hypothesis of non-stationarity is rejected, and it is concluded that the time series is stationary. Using our time series data, here is an example of what we are witnessing in our time series data set prior taking lagged differences: Table 10.1: ADF Results cme open cme close dce open dce close brz open brz close 0.6740813 0.6526461 0.7188882 0.7257436 0.6281175 0.6078181 From the above results, we see that the p-values are well above 0.05, hence we cannot reject the null hypothesis that the time series data is non-stationary. This prompts us to execute further analysis using lagged differences. In the next case, I have used one period lagged difference for the ADF test. Table 10.2: ADF Results cme open cme close dce open dce close brz open brz close 0.01 0.01 0.01 0.01 0.01 0.01 After taking the differences, we see that the p-value is indeed lower than 0.05, which allows us to reject the null hypothesis. Therefore, the time series is stationary after taking first order differences. 56 10.3 Kwiatkowski-Phillip-Schmidt-Shin test (KPSS test) KPSS test is also used in combination with ADF test. It is similar to ADF, how- ever, the null hypothesis is the opposite of ADF, and has added benefits to ADF. The KPSS regression equation: yt = µt + εt where: • yt represents the observed time series data. • µt is the deterministic trend component. • εt is the stationary error term. Null hypothesis: series is trend stationary or series has no unit root Alterna- tive hypothesis: series is non-stationary, or series has a unit root When we have a p-value that is larger than the chosen significance level, which is usually 0.05, then we do not reject H0 and we can assume stationarity. The KPSS test statistic is computed as the sum of the squared residuals from the regression equation: KPS S = T · T∑ t=1 ε̂2 t σ2 where: • T is the number of observations in the time series. • ε̂t represents the estimated residuals from the regression equation. • σ2 is the variance of the estimated residuals. 57 Under the null hypothesis, the KPSS test statistic asymptotically follows a non-standard distribution. Critical values for different significance levels are available in statistical tables. There are other differences with ADF besides the hypothesis test. KPSS tests for both trend and difference stationarity, while ADF tests only difference sta- tionarity. Difference stationarity means that differencing our series will make it stationary. Here are the results of our data series prior differencing: Table 10.3: KPSS Results cme open cme close dce open dce close brz open brz close 0.01 0.01 0.01 0.01 0.01 0.01 All of the above results have p-values below 0.05, we reject the null hypoth- esis, hence we can say that the data series are non-stationary. Then I repeated the same test after taking the first-order differences. Table 10.4: Estimation results for equation DCE.Open cme open cme close dce open dce close brz open brz close 0.1 0.1 0.1 0.1 0.1 0.1 Now we see KPSS test results above our 0.05 critical value threshold. This means that we cannot reject the null hypothesis; hence, the data series is station- ary. 58 CHAPTER 11 RESULTS 11.1 Using First Order Differences (FOD) Model in log form Now that we have established the theoretical and statistical foundations of our model, I am going to present the results for each trade war period from our first order differences model in log-log form. 11.2 Signaling between CME and DCE markets 11.2.1 Prior TW Table 11.1: Estimation results for equation DCE.Open Estimate Std. Error t value Pr(> |t|) CME.Close.l1 1.067e-01 6.041e-02 1.767 0.0779 . DCE.Open.l1 -1.958e-01 4.584e-02 -4.271 2.36e-05 *** CME.Close.l2 1.002e-01 6.047e-02 1.657 0.0981 . DCE.Open.l2 -2.263e-02 4.576e-02 -0.494 0.6213 const -9.479e-05 8.053e-04 -0.118 0.9064 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 0.01753 on 469 degrees of freedom Multiple R-Squared: 0.04405, Adjusted R-squared: 0.0359 F-statistic: 5.403 on 4 and 469 DF, p-value: 0.0002925 59 Table 11.2: Estimation results for equation CME.Open Estimate Std. Error t value Pr(> |t|) CME.Open.l1 2.103e-02 4.715e-02 0.446 0.6558 DCE.Close.l1 8.724e-02 3.860e-02 2.260 0.0243* const 3.206e-06 5.651e-04 0.006 0.9955 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 0.01231 on 472 degrees of freedom Multiple R-Squared: 0.01289, Adjusted R-squared: 0.008709 F-statistic: 3.082 on 2 and 472 DF, p-value: 0.04679 The p-values for the BTW period are statistically significant for both markets. For DCE market in China, CME closing prices come to be significant at both lags. This shows that U.S. prices are a strong indicator of future prices in China. For CME in the U.S., we see that their opening prices are more predicted by the prices in DCE, China than by its own yesterday’s price. The results are in accordance with economic theory, where we hypothesized that price signalling will be strong and statistically significant both ways prior the trade war. 11.2.2 TW Analysing the TW period once China enforced its tariffs on U.S. soybeans in July 2018, we see that signalling has been disrupted in both markets and the prices no longer follow each other as before. For DCE opening prices, we see that both CME lags lost their statistical significance. While for CME, in addition to DCE closing prices, also its own previous opening prices were not a good predictor 60 Table 11.3: Estimation results for equation DCE.Open Estimate Std. Error t value Pr(> |t|) CME.Close.l1 0.0829272 0.0543104 1.527 0.128 DCE.Open.l1 -0.2635019 0.0565219 -4.662 4.66e-06 *** CME.Close.l2 0.0813089 0.0537904 1.512 0.132 DCE.Open.l2 -0.0291834 0.0556375 -0.525 0.600 const -0.0003065 0.0006001 -0.511 0.610 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 0.01063 on 310 degrees of freedom Multiple R-Squared: 0.07568, Adjusted R-squared: 0.06375 F-statistic: 6.345 on 4 and 310 DF, p-value: 6.42e-05 Table 11.4: Estimation results for equation CME.Open Estimate Std. Error t value Pr(> |t|) CME.Open.l1 -0.0593035 0.0561564 -1.056 0.292 DCE.Close.l1 0.0646220 0.0702970 0.919 0.359 const 0.0004499 0.0006626 0.679 0.498 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 0.01176 on 313 degrees of freedom Multiple R-Squared: 0.005143, Adjusted R-squared: -0.001214 F-statistic: 0.8091 on 2 and 313 DF, p-value: 0.4462 of CME opening prices. This may mean that soybean tariffs introduced a shock accompanied by volatility, which made the prices less predictable than before the trade war. 61 11.2.3 Post TW Table 11.5: Estimation results for equation DCE.Open Estimate Std. Error t value Pr(> |t|) CME.Close.l1 0.0760669 0.0394848 1.926 0.05443 DCE.Open.l1 -0.1320756 0.0368457 -3.585 0.00036 CME.Close.l2 0.0065179 0.0395709 0.165 0.86921 DCE.Open.l2 0.1030204 0.0367269 2.805 0.00516 const 0.0005091 0.0006198 0.821 0.41165 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 0.01676 on 730 degrees of freedom Multiple R-Squared: 0.03636, Adjusted R-squared: 0.03108 F-statistic: 6.886 on 4 and 730 DF, p-value: 1.92e-05 Table 11.6: Estimation results for equation CME.Open Estimate Std. Error t value Pr(> |t|) CME.Open.l1 -0.0217226 0.0374249 -0.580 0.562 DCE.Close.l1 0.0388817 0.0399003 0.974 0.330 CME.Open.l2 0.0382277 0.0374022 1.022 0.307 DCE.Close.l2 -0.0096096 0.0399128 -0.241 0.810 const 0.0006357 0.0005927 1.073 0.284 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 0.01602 on 730 degrees of freedom Multiple R-Squared: 0.003131, Adjusted R-squared: -0.002331 F-statistic: 0.5732 on 4 and 730 DF, p-value: 0.6822 62 For the PTW period after the phase one agreement, we see interesting results. For DCE, we see that CME became statistically significant at the first lag. It in- dicates that DCE opening prices are again reacting to closing prices from CME, since China started to import more soybeans from the U.S.. On the other hand, we see the unusual no statistical significance for CME opening prices from DCE closing prices the previous day. When we look back at the figure 9, we see that in 2020 DCE soybean futures prices deviated significantly away from those in the U.S. and Brazil. I contribute this to the Covid pandemic and I believe that is mudding the result. Below, I am going to run the test for PTW period again, but rather than starting the period in January of 2020 when the pandemic began impacting China, I am going to start the PTW period a year later, in January of 2021. Below are the results for both markets: Table 11.7: Estimation results for equation DCE.Open Estimate Std. Error t value Pr(> |t|) CME.Close.l1 0.0813981 0.0276491 2.944 0.00339 ** DCE.Open.l1 -0.0436948 0.0437733 -0.998 0.31865 const -0.0003561 0.0004731 -0.753 0.45198 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 0.01076 on 515 degrees of freedom Multiple R-Squared: 0.01775, Adjusted R-squared: 0.01393 F-statistic: 4.653 on 2 and 515 DF, p-value: 0.00994 63 Table 11.8: Estimation results for equation CME.Open Estimate Std. Error t value Pr(> |t|) CME.Open.l1 -0.0574580 0.0452553 -1.270 0.2048 DCE.Close.l1 0.1689342 0.0845736 1.997 0.0463 * const 0.0001619 0.0007608 0.213 0.8316 Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 Residual standard error: 0.0173 on 515 degrees of freedom Multiple R-Squared: 0.008913, Adjusted R-squared: 0.005064 F-statistic: 2.316 on 2 and 515 DF, p-value: 0.09972 The results above seem to confirm my suspicion that the pandemic altered prices on world markets, especially in China. Hence, once I adjusted the time period to start in 2021, once the fear and shock of pandemic settled down, we see that the prices on both markets started to follow each other again. DCE opening prices for soybeans are more predictable by CME closing prices than its own previous opening price. The same goes for CME opening prices, where DCE closing prices are a better predictor than previous CME opening price. Figure11.1 summarizes p-value results for signalling between CME and DCE markets. Blue is for signaling p-values coming from CME to DCE markets. Red is for signaling p-values from DCE to CME markets. It highlights the TW period, where p-values became insignificant at 10 percent threshold and then suddenly dropped after the Phase One Deal agreement. The table above uses the values for the Post TW after the first Covid lockdowns starting Jan 2021. 64 Figure 11.1: P-Values for CME and DCE signaling 11.3 Signaling between B3 and DCE markets Now that I analysed the price relationship of soybeans between the U.S. and China, I am going to look at the relationship between the soybean prices in China and Brazil. 65 11.3.1 Prior TW Table 11.9: Estimation results for equation DCE.Open Estimate Std. Error t value Pr(> |t|) BRZ.Close.l1 0.13166 0.05890 2.236 0.025857* DCE.Open.l1 0.78673 0.04543 17.318 < 2e − 16*** BRZ.Close.l2 -0.09818 0.05940 -1.653 0.099040. DCE.Open.l2 0.16687 0.04507 3.703 0.000239*** const 0.04924 0.03795