An Alternative Auction Model in Ride Sharing Platforms
(co authored with Ivan Jia)
Executive Summary
We are all familiar with the way Uber and Lyft works. Riders input a destination and the platform provides an estimate of the ride’s cost, followed by sending a request to available drivers. During peak hours, where demand outstrips supply, both Uber and Lyft applies surge pricing which is algorithmically determined by the existing demand curve for rides. Though this process simplifies the user experience as how we are familiar with it, in this short paper we will explore how Didi Dache (滴滴打车), the chinese equivalent of Uber at its inception applied an alternative model. This alternative model allowed users to bid in fixed increments on top of the base ride cost, increasing the likelihood of getting a ride when demand outstrips supply. As this model was eventually ceased as a result of government intervention, we analyzed the rationale of using this model in the first place and seek to explain the underlying mechanisms that made this operating model successful.
The User Experience
After the rider has requested for a ride, the following screen (translated) will be displayed to indicate surge pricing.
Riders are recommended an amount to tip, and are shown the likelihood of them being able to obtain a ride in this current market condition. The probability given allows the rider to assess their own payoffs, that is their total willingness to pay, and decrease their bids in proposed step sizes, in line with descending ride probabilities.
Principles of the Model
The goal of Didi Dache was to use data to precisely match supply and demand to generate more optimized products and services. The company uses a customer centric approach to build this process, taking into considerations the inefficiencies of the existing physical ride hailing model.
On the demand side, when the need for a cab arises, standing alongside the road in attempt to flag down a cab results in a negative payoff for the rider that increases with wait time. On the supply side, some drivers might want to avoid congested roads, or inconvenient routes they might have to take.
By using existing data, this model accounts for the psychology of consumers, and suppliers can omit direct negotiations on the intermediate links. In an example case, for an inconvenient route, if most of the passengers add an extra $20, some $30 and others $10, then the platform draws $21. This functions as a bid for appropriate bargaining of prices between supply and demand.
Network theory explanation:
In an ideal scenario, we then constructed a nash equilibrium in the game theory table(s) to illustrate the payoffs we can observe on both the demand and supply side. Table 1 illustrates the perspectives from the driver and Table 2 the rider.
The driver is operating in an open auction with descending bids and does not have clear visibility over the existing demand situation. While exercising their own judgements, there are two possible scenarios, where supply is greater or less than demand. At this time, if the driver does not respond immediately, they may potentially lose their income as a form of sunk cost at the reserve price set by the platform.
The rider on the other hand also operates in a different auction market: a sealed first price bid auction. Riders have limited visibility of the demand market using only the hail success probability provided, and they can then assess their own payoff and increase their bids, matching the sunk cost of the drivers.
Riders have no clue as to how much others are actually bidding as the hail success probability is built on historical demand. The only confirmation they get is when drivers accept their order. Here, we can intuitively assume that unless they are sure of their payoff amounts, they are not incentivized to increase their bids and will bid the minimum to acquire a positive payoff. In reality, riders tend to increase their fares gradually in this open market situation.
Conclusion
This unconventional model presents a unique case where markets are expected to reach effective clearing prices organically using different auction models on either sides. Despite the model’s initial outlook that it seems to make logical sense, we can observe a few weaknesses that may have potentially caused government intervention.
The first weakness is the shifting of power to riders, the demand side, and resulting in an imbalance of network forces. Earlier, we discussed how drivers have to exercise judgement as they have no visibility over the current demand situation and they might lose their sunk cost. This causes a push back from drivers, incentivizing drivers to form a coalition and artificially create demand.
The second weakness stems from the first weakness, best described as a lack of transparency in the efforts to bring both supply and demand side to a natural market clearing situation. The root cause is the use of two different kinds of auction models on both sides that actually incentivizes either side to form a coalition. In an hypothetical example, riders like drivers, could themselves form an open network and decide not to collectively bid higher.
The third weakness is the lack of a third party, in the case of Uber and Lyft, the network to mediate these transactions. Without a central authority to regulate matching, the network effects become more vulnerable to exploits.
Currently, the Uber and Lyft approach seems better suited for the ride hailing marketplace. We believe that it works because it doesn’t just overcome the three weakness presented, but as the shift of power is passed to the network with charges more upfront and transparent, this also allows the two largest players to compete fairly for the larger share of the ride hailing market. This open competition incentivizes the platforms to innovate, improve at the benefit of both users and drivers, at the expense of multi homing costs.
Sources
[1] http://tech.sina.com.cn/zl/post/detail/i/2017-02-13/pid_8509835.htm?cre=zl&r=user&pos=4_4
[2] http://news.sina.com.cn/o/2017-01-22/doc-ifxzutkf2327804.shtml
