A Study On Supply Risk Management Using Approximate Dynamic Programming

In recent years, supply chains have become increasingly globalized. As a conse-quence, the world’s supply of all types of parts has become more susceptible to disrup-tions. Some of these disruptions are extreme and may have global implications.Our research is based on the supply risk management problem faced by a manu-facturer. We model the problem as a dynamic program. To overcome the well-knownthree curses of dimensionality of dynamic programming, we design and implement ap-proximate dynamic programming (ADP) algorithms to solve the problem. Using nu-merical experiments, we compare the performance of diferent ADP algorithms. Afterselecting algorithm and key parameters, we design a series of numerical experimentsto study the performance of diferent sourcing strategies under diferent settings, tryingto find some insights for supply risk management practice.The results of diferent ADP algorithms show that, compared with Markov deci-sion processes (MDP), ADP can obtain reasonable solutions in much less solution time;compared with myopic algorithm, ADP can obtain more accurate solutions. ADP algo-rithm around pre-decision state variable converges faster than ADP algorithm aroundpost-decision state variable in early iterations, but ADP algorithm around post-decisionstate variable performs better overall with less solution time.Using systematic numerical experiments, we study single sourcing, dual sourcing,multiple sourcing and contingent sourcing strategies. The results of diferent sourcingstrategies show that the addition of a third supplier brings insignificant benefits undera wide variety of settings. Thus, companies in practice can limit their analysis of alter-natives to choosing between a backup supplier (contingent sourcing) and an additionalregular supplier (dual sourcing). Our results suggest that, whether the backup supplierhas lead time or not has significant impact on the performance of contingent sourcing.