Cruise dynamic pricing based on SARSA algorithm

It is a common practice to promote highly discounted fares by cruise companies to enlarge the market share, ignoring economically sustainable development. In some regions, the continuous discounted fares leading to the unsatisfying revenue may be the main cause of decline in ports calls. Cruise companies have learned that dynamic pricing would be much more advantageous at revenue management instead of blindly lowering fares. This paper illustrates such an attempt. We try to dynamically price multiple types of staterooms with various occupancies and evaluate the effect on demand and revenue from different discount and refund policies. We first formulate the cruise pricing problem as Markov Decision Process and Reinforcement Learning (RL), more specifically, state-action-reward-state-action (SARSA) algorithm, is applied to solve it. We then use empirical data to validate the feasibility of RL. Results show that both revenue and demand could be improved under reasonable discount policies. In addition, we demonstrate that reasonable refund policies can also facilitate revenue growth. Finally, a comparison between SARSA algorithm and Q-learning algorithm is discussed. Our finding suggests that SARSA results in higher revenues but takes more time to converge.