Time Series Forecasting of US Inbound Border Crossings by Commercial Trucks

This study was designed to assess how Customs and Border Protection could become more efficient and effective by informing short-term staffing, medium-term budgeting, and long-term investing decisions with forecasts of monthly inbound commercial truck border crossings. To do this, the study reviewed the relevant literature to guide the model-building approach, built and evaluated several candidate models, and assessed how the model forecasts could be incorporated into CBP operations and inform decision-making. The models employed were ARIMA, REGARIMA, OLS, LASSO, Random Forest, XGBoost, and Cubist. Although no model or forecast is perfect, and there is inherent risk in basing key decisions on forecasts, it seems likely that CBP can benefit from the models, provided they diligently consider the balance and tradeoffs of competing priorities, and the flexibility, cost, and risk of errors given the particular context for each decision.