An ambiguous manager's disruption decisions with insufficient data in recovery phase

In this paper, we study the manager's decisions in mitigating the disruption of the operation system when its operating units (OUs) were crippled by the unexpected event. The manager's ambiguity in making decisions during the recovery phase might generate two results: first, the level of remaining OUs might fluctuate due to uncontrollable ripple effects, and second, the lack of sufficient historic disruption could create unanticipated consequences. In this study, the ambiguous manager is described as the Choquet, one whose ambiguity belief is represented by the Choquet expect utility. Through the single- and multi-periods recovery models, we show that the Choquet manager consistently procures more short-term OUs from his capacity-shared partners than the rational one, and tends to build a higher redundant inventory level in the pre-disruption phase. To investigate the impact of insufficient demand data on the decisions of the Choquet manager in the recovery phase, we adopt two Bayesian learning methods to dynamically update the ambiguity belief in a multi-periods setting: first, the Beta method (a parametric method), and second, the minimum relative entropy (MRE) method (a nonparametric, also a data-driven method). Numerical results present findings in three aspects: First, the MRE method provides more robust estimations than the Beta one, and hence, it leads to a lower disruption cost because of its better approximation to the distribution of the uncertainty. Second, the initial redundant inventory does not contribute as much to lower the disruption cost as shortening the recovery time. Third, there is an “anchoring effect” when the manager's follow-up decisions are anchored on previous estimations of the the uncertainties’ mean value.