The Design of Centralized Matching Systems on Two-Sided Platforms: Evidence from the Ride-Hailing Market
Preface: The “Ghost Cars” of London
Working in London, I often find myself staring at a ride-sharing app showing cars just minutes away, yet spending over 10 minutes waiting for a match. The cars are there, but they are ghosts—visible, yet inaccessible.
One driver explained the paradox to me: “We are picky.” He admitted to ignoring short trips or unpopular destinations to wait for “good” runs like airport trips. However, he also shared a moment of doubt: “Honestly? I spend so much time sitting around rejecting rides… sometimes I feel like I’m earning less money than if I just took whatever popped up.”
That driver’s intuition highlights the central tension of the gig economy: Autonomy vs. Efficiency. Does the freedom to choose actually help workers, or does the collective friction of everyone “cherry-picking” leave everyone—drivers and riders alike—worse off?
This research project was born from that question.
The Dilemma: To Choose or To Assign?
In the platform economy, there are two dominant ways to match supply and demand:
- Driver-Accept (Agent Choice): The platform broadcasts a request to nearby drivers. Drivers can accept or ignore it based on their preferences. (Used by Uber in many markets, HKTaxi, BidRide).
- Auto-Accept (Platform Assignment): The platform algorithmically assigns a trip to a specific driver. The driver generally cannot decline without penalty. (Used by Grab, DiDi for short trips, and Lyft in some modes).
The industry is divided.
- Proponents of Choice argue that drivers treat their work as a business. They possess “private information”—they know when they are tired, which areas feel unsafe, or if they need to head home soon. Allowing them to choose respects their autonomy.
- Proponents of Assignment argue that choice creates friction. If every driver waits for a “unicorn” trip, utilization drops. Customers wait longer. Reliability plummets.
We set out to settle this debate empirically.
The Study
We partnered with a leading taxi operator in Singapore. During our study period, the company used a Driver-Accept system. When a customer requested a ride, it was broadcast to multiple nearby drivers. Drivers could see the details and decide whether to bid.
This setting was perfect for our analysis because it allowed us to observe strategic rejection. We built a dynamic structural model of this two-sided market.
- The Riders: Make choices between street-hailing, e-hailing, or outside options based on wait times and prices.
- The Drivers: Are forward-looking strategic agents. They don’t just maximize the current fare; they think about where a trip takes them. “If I go to Changi Business Park at 5 PM, I’ll get a great trip back. If I go to Tuas, I’m stuck.”
Using granular GPS and transaction data, we estimated this model to understand the deep preferences and strategic behavior of drivers. Then, we ran a counterfactual simulation: What if the platform switched to an Auto-Accept system?
What We Found
The results were counter-intuitive to many drivers, but validated the hunch of my London driver.
1. Removing Choice Increased Driver Earnings
In our simulation, switching to an Auto-Accept system increased the average driver’s earnings.
Why? The Utilization Effect. Under the Driver-Accept system, drivers spent a significant amount of time empty, waiting for a “better” offer. They were over-optimizing. By rejecting average trips in hopes of a great trip, they incurred too much idle time. The Auto-Accept system eliminated this search friction. Drivers were kept moving. The volume of trips they completed compensated for the inability to cherry-pick high-value individual rides.
2. The “Rising Tide” Effect (Long-Term Equilibrium)
The benefits compounded over time.
- Better Service: Because drivers couldn’t reject “bad” trips, acceptance rates went up.
- Lower Wait Times: Riders got cars faster.
- More Demand: Because the service became more reliable, more people started using the app.
This created a virtuous cycle. The market expanded, further increasing driver utilization and earnings.
3. Consumer Surplus Increased
Unsurprisingly, customers were also winners: discriminatory rejection of trips disappeared. Wait times stabilized. The reliability of the service improved significantly.
Conclusion
Our research suggests that in high-density, real-time matching markets like ride-hailing, centralized assignment can beat decentralized choice.
While “autonomy” is a powerful value proposition for the gig economy, it comes with hidden costs. Search frictions—the time spent looking for the perfect match—are a deadweight loss for everyone. By moving to an algorithmic assignment system, platforms can actually make drivers better off financially, even if it feels like they are losing control. Ideally, platforms should find ways to compensate for the psychological loss of autonomy while delivering the economic gains of efficiency.
For my London driver, the answer turns out to be yes: if the system just told him where to go, he likely would have made more money that night—and I would have gotten home 10 minutes earlier.