Analyzing U.S. Maritime Trade and COVID-19 Impact Using Machine Learning

Introduction

Americans import and export goods in three primary ways: air, land (truck, rail, and pipeline), and sea. Out of the three methods, maritime trade is the primary vessel of international trade. As of 2013, as much as 53% of imports and 38% of exports move through American ports, measured in U.S. dollars. The deepening of the Panama Canal and local passages around U.S. ports has encouraged ocean carriers to realize economies of scale by using larger vessels. The maximum size of a ship increased from about 15,000 twenty-foot equivalent units (TEUs) in 2010 to 23,000 TEUs in 2020 (Manaadiar, 2020).

Although monetary terms are one form of measuring value, its meaning is relative to the current exchange rate and value of the USD. Economists have traditionally focused on monetary measurements such as the product's value and national GDP. Instead, we opt to measure trade in metric tonnage. Over time, general baskets of goods retain relatively constant weights whereas real values vary based on exchange rates, scarcity, and inflation. When measured in metric tonnage, maritime trade accounts for 72.4% of imports and 74.8% of exports to and from the United States. In 2019, the Bureau of U.S. Customs and Border Protection cleared 11,160,342 shipments. Serious importers and exporters overwhelmingly prefer sending their goods by sea.

Goods can be transported by sea in a variety of ways, including liquid bulk, breakbulk, roll-on/roll-off (ro/ro), and containers, with the latter being the preferred method for most shippers (Mittal, Boile, Baveja, & Theofanis, 2013). Liquid bulk can be measured in ISO tanks, ro/ro in units, and breakbulk in cartoons, bags, or boxes. The variety of measurements can make cross-comparison difficult. This is yet another reason to use metric tonnage as the standard unit of comparison. Although valuations in the U.S. dollar may facilitate understanding in broad economic terms, the analysis is only useful for shorter time intervals, when framed in the specific time period studied.

This paper begins by elaborating on literary, industry, and historical support that justifies the chosen attributes and time period. Next, details related to data collection and methodologies are outlined. Data analysis and results are divided into subcategories for each model. Models are developed and evaluated to answer the research questions. This is followed by an analytics section to highlight trends. We conclude with a section with recommendations for future research.