考量耐心消費者與參考效應之動態定價

Abstract

隨著網路發展及比價工具的普及，消費者不再是被動的價格接收方。消費者因價格偏高而延遲購買的耐心行為，讓定價決策變得更加複雜，對動態定價構成挑戰。如果業者未將延遲購買行為納入定價考量，可能會錯失最佳定價，導致整體收益下降。因此，本研究提出一種結合耐心消費者行為與參考效應的動態定價策略。建立一個可以在未知環境中，進行精準預測和即時決策的訂價系統。 本研究主要聚焦於航空機票市場。我們運用多項羅吉特模型（MNL)來量化參考價格、市場參考價值及競爭者價格對購買行為的影響，並將其納入動態規劃（DP）模型中以求解最佳定價。此外，我們應用長短期記憶神經網絡（LSTM)來預測消費者耐心行為與價值參數。並透過自適應訓練持續提升模型預測效能，使其能適應市場變動，動態修正預測結果。 為克服動態規劃面臨的維度詛咒問題，我們進一步訓練深度神經網絡（DNN）以取代動態規劃，使其能夠即時做出精準的定價決策。本研究同時設計數值模擬，以測試模型在未知市場中的適應能力。模擬結果顯示，若能精準預測消費者的耐心程度，將能有效提升收益。本研究模型在市場中，能維持穩定效能，並展現良好的應用價值，可供時效性商品業者參考。

With the development of online markets and the widespread availability of price comparison tools, consumers are no longer passive price takers.Patient consumers tend to delay purchases when prices are perceived as high, which complicates pricing decisions and poses challenges for dynamic pricing. If firms fail to account for such delayed purchasing behavior, they may miss optimal pricing opportunities and experience reduced overall revenue. To address this issue, this study proposes a dynamic pricing framework that integrates consumer patience and reference price effects, enabling accurate prediction and real-time pricing decisions under uncertain environments. This research focuses on the airline ticket market. A multinomial logit (MNL) model is employed to quantify the influence of reference price, market reference value, and competitor's prices on consumer choice, and these factors are incorporated into a dynamic programming (DP) model to derive optimal prices. In addition, long short-term memory (LSTM) neural networks are used to predict consumer patience levels and demand parameters. An adaptive training mechanism is further applied to continuously improve prediction accuracy, allowing the model to adjust to market changes and dynamically refine its estimates. To overcome the potential curse of dimensionality in dynamic programming, a deep neural network (DNN) is trained to approximate the DP solution and generate real-time pricing decisions. A discrete-event simulation is also developed to evaluate the model's adaptability in previously unseen market environments. Simulation results show that accurate prediction of consumer patience significantly enhances revenue performance.Overall, the proposed model demonstrates stable performance and strong practical value,providing a useful reference for industries dealing with perishable or time-sensitive products.