Thinking outside the container: A sparse partial least squares approach to forecasting trade flows
利用全球集装箱船运动数据构建时间序列，结合LASSO和偏最小二乘法预测月度贸易流量，提前数月获得优于基准模型的预测结果，对76个国家和地区均有效。
Global container ship movements may reliably predict trade flows. First, this paper provides the methodology to construct maritime shipping time series from a dataset comprising millions of container vessel positions annually. Second, to forecast monthly goods trade using these time series, this study outlines the use of the least absolute shrinkage and selection operator (LASSO) in combination with a partial least squares process (PLS). An expanding window, out-of-sample exercise demonstrates that constructed forecasts outperform benchmark models for the vast majority of 76 countries and regions. The performance holds true for unilateral and bilateral trade flows, for trade of developed and developing countries, for real and nominal trade, as well as for time periods of economic crisis such as the COVID-19 pandemic. The resulting forecasts of trade flows precede official statistics by several months and may facilitate quantification of supply chain disruptions and trade wars.