TL;DR
This paper introduces a novel agent-based simulation framework, MaaSSim, to model the complex dynamics of two-sided mobility markets (like ride-sharing). It uses an S-shaped learning model to capture how individual drivers and riders adapt their behavior based on experience, word-of-mouth, and marketing, successfully reproducing the rise and fall of such platforms under different market entry strategies.
What Problem Does This Paper Address?
Two-sided mobility platforms, such as Uber and Lyft, have experienced rapid growth but also face challenges in achieving profitability and stability. Understanding the intricate interplay between user behavior, platform strategies, and market dynamics is crucial for predicting their success or failure. Existing models often struggle to capture the day-to-day evolution of these markets, particularly the nuanced adaptation of individual agents. This paper aims to fill this gap by developing a microscopic, agent-based model that realistically simulates how agents learn, adapt, and make decisions, thereby reproducing the full lifecycle of a two-sided mobility market, including its rise and eventual fall.
Methods
The core of this research lies in the development of a novel co-evolutionary agent-based model implemented within the MaaSSim simulator. The key innovation is the use of an S-shaped learning function to update agents' perceived utility. This function models how agents' utility is influenced by three components:
- Experienced utility: Derived from direct interactions with the platform (e.g., travel times, driver income).
- Word-of-mouth (WOM) utility: Reflecting social influence from peers.
- Marketing utility: The impact of platform advertising.
Unlike previous approaches that might use exponential smoothing, the S-shaped curve better captures how agents' perceptions stabilize over time, yet remain sensitive to significant changes. Agents are modeled as rational utility maximizers who adjust their behavior daily. The model also incorporates a realistic road network (Amsterdam) and a six-stage market entry strategy for the platform, including phases like "kick-off," "discount," "launch," "growth," "maturity," and "greed."
Datasets
The study uses a simulated dataset based on the road network of Amsterdam. Trip data is sampled from the real-world Albatross tripset, filtered for trips longer than 2 km to ensure empirical plausibility. Public transport data is obtained using OpenTripPlanner on a detailed OSM graph and GTFS timetable. Driver reservation wages are based on the minimum daily wage in the Netherlands.
Key Insights and Ideas
- S-shaped learning is crucial: The S-shaped utility update mechanism is vital for realistically modeling agent adaptation and market stabilization, outperforming simpler models.
- Marketing as an initial trigger: Marketing campaigns are essential for initially notifying agents and incentivizing them to try the platform, especially in the early "launch" stage.
- WOM and experience sustain growth: After the initial marketing push, positive word-of-mouth and accumulated positive experiences become the primary drivers of sustained growth.
- Discounts are effective but costly: While discounts can boost market share, they lead to initial losses for the platform, which are only compensated by commission fees later.
- Greed leads to collapse: An aggressive strategy, like drastically increasing commission fees (the "greed" stage), can quickly trigger negative network effects and lead to a catastrophic loss of market share.
- Individual heterogeneity matters: Each agent follows a unique evolution path based on their specific exposure to marketing, WOM, and personal experiences, highlighting the importance of microscopic modeling.
Key Results
The simulation successfully reproduces a six-stage market evolution over 400 days:
- Kick-off & Discount (Days 0-50): Initial low adoption despite discounts, as agents are not yet aware of the platform.
- Launch (Days 50-100): Marketing campaign drives significant user acquisition, reaching ~20% market share.
- Growth (Days 100-200): Sustained growth fueled by WOM and positive experiences, leading to ~60% market share.
- Maturity (Days 200-300): Platform stabilizes with ~60% market share after discounts are removed.
- Greed (Days 300-400): A sharp increase in commission rate to 50% causes a dramatic collapse in market share for both drivers and travelers, leading to significant losses.
The model demonstrates that platform profitability is achieved around day 250, after compensating for marketing costs. The "greed" stage, while initially boosting profit, ultimately leads to a decline due to lost market share.
Limitations and Future Work
The authors acknowledge that the model uses a synthetic uniform probability for WOM diffusion, which could be replaced with the actual structure of social networks for more granular analysis. They also note that heterogeneous preferences could be further explored by setting different utility weights for agents.
Future work could involve applying this framework to various research problems, such as analyzing minimal wages, platform competition, congestion charging, driver number caps, ride-pooling, and even the lobbying efforts of ride-sourcing platforms. The model's ability to capture individual agent trajectories also opens avenues for studying adaptation processes based on specific preference-genes.
Summary updated