Causal modeling of Incoterms® selection using directed acyclic graphs and machine learning: evidence from Colombian customs

Data for the paper "Causal modeling of Incoterms® selection using directed acyclic graphs and machine learning: evidence from Colombian customs"<br><b>Abstract</b>:<br><br>In international trade, selecting an Incoterm is a critical decision that directly impacts a firm's costs and risks. Previous quantitative studies have relied on traditional econometric models with restrictive assumptions, which limit their ability to provide causal insights for decision-making. This paper addresses this gap by developing and applying a causal inference framework to empirically evaluate the key determinants of this choice for Colombian exporting firms. Our methodology integrates directed acyclic graphs to explicitly model causal assumptions, uses nested cross-validated random forest models to capture nonlinear relationships, and employs the G-computation formula to estimate robust causal effects. Our results show that specific transactional characteristics primarily influence Incoterm selection. For example, shifting a shipment from bulk to containerized cargo increases the likelihood of selecting an E/F group term (e.g., FOB or FCA) by 13.4 percentage points. We also identify nonlinear causal effects for shipment weight and value. E/F terms become more probable only after reaching a certain size threshold, a pattern not identified by prior linear models. Having a common language with the trade partner also has a significant causal effect, increasing the probability of using E/F terms by 6.5 percentage points. This research provides a data-driven framework for strategic Incoterm selection, enabling managers to align their choices with specific shipment characteristics. The findings also offer practical recommendations for logistics providers designing targeted services and for policymakers developing effective trade support programs.<br><br><br><br>