Abstract
In urban mobility services, e-coupons are frequently used to manage demand and foster loyalty. However, empirical evidence suggests that travelers often fail to redeem valid coupons even when it is financially optimal. This paper proposes an "inattention model" that incorporates traveler unawareness into a Markov Decision Process (MDP) framework. By modeling the "attention recovery" that occurs when a traveler opens their digital wallet, the study provides a more accurate representation of real-world behavior. Analysis of a large-scale car-sharing dataset from Shanghai shows that the inattention model significantly outperforms traditional utility-maximization models, revealing that traveler forgetfulness can diminish promotional effectiveness by approximately 10%.
Introduction
Standard economic models assume that "rational" agents possess perfect information and always act to maximize their utility. In the context of ride-sharing or car-sharing, this implies that a traveler should always use an available e-coupon if the trip fare exceeds the coupon's value. Yet, data consistently shows a gap: many travelers pay full price while discounts remain unused in their digital wallets.
This paper addresses this discrepancy by identifying unawareness as a primary behavioral barrier. Travelers do not necessarily ignore coupons; rather, they often forget they exist until they are prompted to interact with their digital "e-wallet."
Methodology: Modeling Inattention
The researcher develops a framework based on a Markov Decision Process (MDP) to capture the sequential nature of travel decisions. While a baseline MDP model assumes perfect rationality—where travelers calculate the "future redemption value" of a coupon—it fails to match observed behavior.
To bridge this gap, the Inattention Model introduces two critical concepts:
- Unawareness Subset: Travelers do not choose from the full set of available coupons, but from a perceived subset.
- Attention Recovery: The model assumes that the act of deciding to use any coupon requires opening the e-wallet. This action "recovers" the traveler's attention, revealing the full set of available coupons for the final selection.
The model employs a Manski-style independent awareness probability for each coupon, which is influenced by the traveler's "activation record" (previous interactions with the coupon). Parameters were estimated using TensorFlow and the Adam optimizer to maximize the log-likelihood of observed trip data.
Empirical Evidence
The study is grounded in a massive dataset from a car-sharing service in Shanghai, China, covering 1.5 million trip records from 160,000 travelers between January and July 2017. The data confirmed the "smoking gun" for inattention: even when fares were higher than coupon values, redemption rates remained significantly below 100%.
Key observations from the data include:
- Complexity Effects: As the number of coupons in a wallet increases, the divergence from rational utility-maximization grows.
- Myopic Behavior: Travelers tend to value immediate discounts far more than potential future savings, with the estimated weight for future coupon value (theta_V) consistently below 1.
Results and Economic Impact
The inattention model demonstrated a significantly better fit for the dataset than the baseline model. In scenarios involving a single coupon, the model achieved a forecasting accuracy of 78-79%.
Beyond predictive power, the model highlights the economic consequences of traveler forgetfulness. Simulation experiments indicate that unawareness leads to a 10% reduction in the promotional effectiveness of coupon schemes. This suggests that "passive" distribution—simply placing a coupon in a user's account—is inefficient. Active reminders and notifications are essential to overcome the unawareness barrier and maximize the impact of marketing spend.
Conclusion and Future Directions
While the inattention model provides a robust framework for understanding traveler behavior, several areas for future research remain. The current model assumes independent coupon-level awareness, which may be overly simplistic for travelers holding many similar coupons. Future work could explore "homogeneity" between coupons or utilize more complex transition dynamics, such as Recurrent Neural Networks (RNNs), to model how awareness fluctuates over time. Ultimately, the study underscores that in the digital economy, managing consumer attention is as critical as managing price.
Summary updated