We study the facility location problem with disruptions where the objective is to choose a set of locations that minimizes the sum of expected servicing and setup costs. Disruptions can affect multiple locations simultaneously and are caused by multiple factors like geography, supply chain characteristics, politics, and ownership. Accounting for the various factors when modeling disruptions is challenging due to a large number of required parameters, the lack of calibration methodologies, the sparsity of disruption data, and the number of scenarios to be considered in the optimization. Because of these reasons, existing models
neglect dependence or pre-specify the dependence structures. Using partially subordinated Markov chains, we present a comprehensive approach that starts from disruption data, models dependencies, calibrates the disruption model, and optimizes location choices. We construct a metric and a calibration algorithm that
learns from the data the strength of dependence, the number of necessary factors (subordinators), and the locations each subordinator affects. We prove that our calibration approach yields consistent estimates of the model parameters. Then, we introduce the precise-cut algorithm, which leverages partially subordinated
Markov chains to solve the resulting optimization problem quickly and precisely. Finally, we demonstrate the efficacy of our approach using three different disruption data sets. Our calibrated parameters are robust, and our optimization algorithm performs better than the simulation-based algorithm. Our approach allows for better modeling of disruptions from historical data and can be adapted to other problems in logistics, like the hub location, capacitated facility location, etc., with joint disruptionly.