A Stackelberg-based deep reinforcement learning approach for dynamic cooperative advertising in a two-echelon supply chain

Dynamic market design through cooperative advertising programs is conducive to generating immediate sales as well as win-win or Pareto-efficient yields. However, in a decentralized supply chain, there is no known optimal advertising and ordering policy for different participants. Motivated by this, we integrate inventory control with the expectation-based reference price in a two-player supply chain where a manufacturer provides a cooperative advertising program for a retailer. After eliciting the problem as an infinite-horizon Stackelberg game, we propose a novel Leader-Follower Deep Deterministic Policy Gradient (LFDDPG) algorithm. Extensive experiments show that our algorithm significantly outperforms metaheuristics with regard to different configurations of demand distribution and costs. Similar results are observed using data from a real-life manufacturer. Furthermore, our algorithm is robust in hyperparameter spaces and exhibits superior convergence behavior. These findings provide valuable insights and pave the way for future research and practical implementations in supply chains with complex decision-making processes.