A budget-balanced and strategy-proof auction for multi-passenger ridesharing

07/17/2020
by   Leonardo Y. Schwarzstein, et al.
University of Campinas
0

Ridesharing and ridesourcing services have become widespread, and properly pricing rides is a crucial problem for these services. We propose and analyze a budget-balanced and strategy-proof mechanism, the Weighted Minimum Surplus (WMS) auction, for the dynamic ridesharing problem with multiple passengers per ride. We also propose and analyze a budget-balanced version of the well-known VCG mechanism, the VCG_s. Under the assumption of downward closed alternatives, we obtain a lower bound for the surplus welfare and surplus profit of the WMS. Finally, we present an exact algorithm based on integer linear programming to solve these auctions. Encouraging experimental results of profit and welfare were obtained for both the WMS auction and the VCG_s.

READ FULL TEXT VIEW PDF
POST COMMENT

Comments

There are no comments yet.

Authors

page 1

page 2

page 3

page 4

04/16/2018

Efficient Auctions With Common Values

Consider the problem of allocating goods to buyers through an auction. A...
04/29/2019

Efficiency in Truthful Auctions via a Social Network

In this paper, we study efficiency in truthful auctions via a social net...
07/18/2017

Distributed Bi-level Energy Allocation Mechanism with Grid Constraints and Hidden User Information

A novel distributed energy allocation mechanism for Distribution System ...
07/17/2019

Envy, Regret, and Social Welfare Loss

Incentive compatibility (IC) is one of the most fundamental properties o...
12/13/2019

Toward an Automated Auction Framework for Wireless Federated Learning Services Market

In traditional machine learning, the central server first collects the d...
07/20/2021

Deterministic Budget-Feasible Clock Auctions

We revisit the well-studied problem of budget-feasible procurement, wher...
12/01/2018

Strategy-Proof Spectrum Allocation among Multiple Operators

To address the demand of exponentially increasing end users efficient us...
This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

1 Introduction

The prevalence of smartphones equipped with internet connectivity and GPS systems have enabled the rapid growth of platforms that allow a passenger to make a short-term request for the service of independent drivers. When the driver’s motivation is to obtain a recurring income from these trips, such as in the platforms Uber and Lyft, this is called ridesourcing (rayle2014app). This scenario can be distinguished from ridesharing, where the driver seeks to share their trip costs by serving passengers that have a route similar to their own, as for example, in Waze Carpool (Waze) — a commercial platform that facilitates ridesharing. Some ridesourcing platforms also seek sharing as a way to reduce costs, for example, UberPool (Uber) and Lyft Line offer a smaller price by allowing the trip to be shared among multiple passengers. The InDriver (InDriver) platform resembles an auction since it allows passengers to advertise rides and prices, to which drivers can then give a counteroffer. These services are part of a new type of economic interaction, called collaborative consumption or sharing economy (hamari2016sharing).

Beyond the benefits for the participants of the trip, ridesharing is advantageous for the society as a whole, as it may reduce the congestion and pollutant emission caused by the use of personal cars in metropolitan areas by increasing the occupancy rate of vehicles and reducing the need for parking (hahn2017ridesharing)

. As an example, the metropolitan area of São Paulo, one of the largest in the world, is in a long-lasting mobility crisis that was estimated to cost up to 1% of Brazilian GDP in 2012 

(cintra2014custos), with congestion resulting in up to three times more emission of pollutants than the expected emission without congestion (cintra2014custos).

In this paper, we propose pricing solutions that apply to both ridesharing and ridesourcing and uses the term ridesharing to generalize both scenarios, except where the difference is explicitly noted. We are particularly interested in dynamic ridesharing with trips that may be shared among multiple passengers, which incurs lower costs for passengers and society. The Dynamic Ridesharing Problem, or real-time ridesharing, is characterized in agatz_optimization_2012 by the following properties: Dynamic, the trips must be formed rapidly as new requests come in continuously; Independent, drivers and passengers are independent agents and therefore will only share a trip if that is beneficial to them; Cost sharing, participants wish to obtain a lower cost than the one incurred by traveling alone; Non-recurring trips, single short-term trips rather than recurring pre-arranged trips; Automatic, the system should require minimal operational effort from the participants.

Dynamic ridesharing involves complex routing and pricing aspects. The Dial-A-Ride Problem (DARP), as defined in cordeau_branch-and-cut_2003, is a good routing model for multi-passenger ridesharing, where passengers have pick-up and drop-off locations that may not coincide with the driver’s initial location and destination while also having constraints on pick up time and travel time. Routing problems that have cost minimization as the objective have received much attention in the operations research literature (pillac2013review), for example, santos2013dynamic

develop heuristics to maximize taxi sharing in a DARP problem. The ridesharing problem with dynamic pricing brings new challenges such as balancing the number of passengers served and the profit obtained.

Trip pricing gains importance in dynamic ridesharing because the independence of the agents makes it necessary to incentivize the participation of passengers and, especially, drivers in the system. However, pricing has received less attention from the ridesharing literature (furuhata_ridesharing:_2013)

. In moments of high demand, when there are few drivers and many passengers, it is necessary to raise prices to attract drivers into the system and balance supply and demand. In this scenario cost minimization is no longer the main goal of the system’s optimization since the valuation of each passenger for the trip must be considered. Some ridesourcing platforms such as Uber and Lyft raise the price per distance in these scenarios, a policy known as

surge pricing. Current studies suggest that surge pricing achieves its goals of improving economic efficiency (hahn2017ridesharing). Still, surge pricing has been viewed negatively by consumers and regulators (cachon_surge). When the system decides for a higher price consumers might perceive that it is acting against them. In an auction however the consumers’ bids would determine the price, so we explore auctions as a form of balancing the market in moments of high demand with a price that is better justifiable to the consumer.

1.1 Literature Review

Two desirable properties of auctions that are studied in auction-based pricing for ridesharing literature are budget-balance, which means the price paid by users is at least the driver’s cost for serving them, and strategy-proofness, which means participants do not gain by misreporting their valuations to manipulate the auction.

kamar_collaboration_2009 propose a multi-driver system based on a Vickrey-Clark-Groves (VCG) (vickrey_counterspeculation_1961; clarke1971multipart; groves_incentives_1973) auction. The VCG auction is not budget-balanced and, due to the use of heuristics for better computational performance, is implemented in a way that also is not strategy-proof. kleiner_mechanism_2011 uses a strategy-proof auction adapted from the second-price auction (vickrey_counterspeculation_1961) where each driver may serve a single passenger. Considering each driver in isolation, this auction is strategy-proof and budget-balanced, as long as the driver is not allowed to choose the served passenger. Our work extends this line of research by proposing mechanisms that are multi-passenger, strategy-proof, and budget-balanced.

Another line of research involves making prior assumptions on the passenger’s arrival rate and valuation distributions and then calculating prices that are optimal in expectation. sayarshad2015scalable and chen_optimal explore this type of mechanism while masoud2017using assumes a prior distribution on the passengers’ valuations to propose an optimal price for which a passenger may buy a trip from another passenger.

In recent literature, ZhaoVCG propose a multi-driver and multi-passenger theoretical model that uses a VCG auction with reserve prices, but they achieve budget-balance only when the drivers make no detours, while the auctions we propose are budget-balanced even with detours. zhang2016discounted apply a bilateral trade reduction mechanism that is strategy-proof, modifying the McAfee mechanism (mcafee1992dominant) for a ridesourcing scenario. A limitation of this mechanism is that a driver may only serve one passenger at a time, and budget-balance is not guaranteed. shen2016online propose an online ridesharing system where a passenger that enters the system is offered a price estimate to be served by an available driver, assuming a scenario where the driver supply is sufficient for passengers to be immediately served. The authors claim that the system is strategy-proof however the final price may be lower than the accepted estimate, so the passenger may strategically accept a higher estimate hoping to get a lower final price. This system is in fact individually rational, that is, it guarantees that a passenger that only accepts estimates lower or equal to their valuation will not suffer a loss.

A comprehensive survey by furuhata_ridesharing:_2013classifies ridesharing as either for a single passenger or multiple passengers. It also describes four spatial patterns of increasing generality, of which the most general is detour ridesharing, where the driver may take detours and passenger pick-up locations do not have to coincide with the driver’s start location and neither do passenger delivery locations have to coincide with the driver’s destination. This survey also identifies challenges for auction-based ridesharing, among them to find a mechanism that is strategy-proof, budget-balanced, and allows multiple passengers to be served. Even in recent literature this challenge still has not been addressed.

1.2 Our contribution

In this paper, we propose auction mechanisms for single-driver, multi-passenger, detour ridesharing that are strategy-proof and budget-balanced. The main proposed auction is the Weighted Minimum Surplus (WMS) auction, for which we obtain a lower bound of the maximum social welfare and profit under certain conditions. We also propose , a budget-balanced version of the VCG auction which remains strategy-proof.

We present an experimental analysis of all proposed auctions, utilizing ridesharing instances generated on real-world maps. Both the WMS and the obtain high social welfare and achieve good profit, particularly in instances with a larger number of passengers. Though these mechanisms were analyzed only in single-driver scenarios, they may serve as a basis for generalization to multi-driver scenarios, which appears to be a much harder problem.

2 Auction-based dynamic ridesharing

In the Auction-Based Dynamic Ridesharing Problem with a Single Driver and Multiple Passengers a driver with a specific destination wishes to share their trip with passengers in exchange for monetary gain. Each passenger wishes to be served, that is, transported from their pick-up location to their drop-off location. To solve the problem using an auction, the system requests each passenger to bid a non-negative value for being served. The passengers have maximum pick-up time and travel time constraints. In a ridesharing scenario, the driver may have a constraint on the maximum arrival time at the destination. We focus on the single-driver problem as a step towards a generalization for the multi-driver problem.

A trip consists of a route that begins at the driver’s current location, serves a set of passengers, and finishes at the driver’s destination, respecting all time constraints. A trip is represented simply by a set of served passengers, that is, if and only if is served by trip . The value  is the cost for the driver to serve , representing costs such as fuel and time. We consider that the distances between locations form a metric, and therefore the triangle inequality is valid. The system must determine the winning trip to be served by the driver and the price to be charged from each passenger. Let be the set of all passengers, then the total value collected will be  (noting that, in our auctions, if ).

An auction is a suitable way to solve this problem when there is a high demand since the price will rise along with the bids as an incentive for drivers to join the system. Since there are multiple desirable properties for an auction studied in Auction Theory, we focus on four properties to evaluate auction mechanisms for this problem: Being strategy-proof, maximizing social welfare, being budget-balanced and maximizing profit. The remainder of this section defines these properties.

The passenger’s bid may not be equal to their private valuation for the ride. Strategy-proofness means that a passenger can maximize their gain by bidding . Therefore, a single passenger cannot manipulate the auction to their advantage by lying their true valuation for being served. Since we restrict our attention to auctions that are strategy-proof we refer to or interchangeably.

Let be set of trips, or alternatives, from which the winning trip is chosen. The social welfare of an alternative is the valuation of the served passengers minus the cost for the driver, that is, . An auction maximizes social welfare if . Budget-balance is achieved when the total price paid is at least the cost for the driver, that is, and the driver never suffers a loss from serving the passengers. Profit is an important goal both for drivers and for the organization running the system and is defined as . To maximize profit is to maximize the surplus payment after paying the driver cost. In the next section, we propose an auction that is strategy-proof, budget-balanced, and achieves good social welfare and profit in experimental results.

3 Weighted Minimum Surplus auction

In this section, we propose the Weighted Minimum Surplus (WMS) auction for a single-driver, multiple passenger setting which models the situation where a driver enters a system where they are the only driver available to multiple waiting passengers. We prove this auction to be budget-balanced, which is desirable to the driver as they will never lose money from a trip, and strategy-proof which is desirable to the passenger who can maximize their gain simply by bidding the true valuation , instead of strategizing to improve their gain.

3.1 Winning trip and price

For the auction to be budget-balanced it must be ensured that the cost of the winning trip will be covered by the prices paid. To achieve this, from the valuation of each passenger , the passenger cost is reserved to pay part of the cost of the driver if is served. The exact way to determine  is a choice to be made by the auction designer. For strategy-proofness to hold it is required that  does not depend on  nor on the choice of winning trip. For example  can be the cost of going from passenger’s pick-up location to their drop-off location.

The surplus of passenger is the difference between the passenger’s bid and the passenger’s cost. Only passengers with can be served by the auction since, otherwise, their valuation would not be sufficient to cover their cost.

A trip will be considered feasible if for all and the passenger costs are sufficient to cover the cost of the trip, that is . The set of alternatives for the auction (from which the winning trip is chosen) is composed only of feasible trips.

Let the minimum surplus of a trip be defined as , then the weighted minimum surplus of is . The winning trip is given by the allocation function  where . Ties are broken by some criteria that does not depend on bids.

To define , first we must define the value and the alternative . Given a passenger  let . If , then is simply the trip of maximum cardinality such that and . In general, is defined for any as the set of maximum cardinality among the sets such that , and

(1)

In case of an equality in Equation (1), must also be such that it wins a tie against any for which . If no such that fulfills all the requirements exists, then if  let . Given and , the price paid by is defined as

(2)

if , and otherwise. By Lemma 1, if , then exists and is well-defined. The term  is used to pay the driver’s cost while the value of the second term may be considered entirely as profit to be shared between the driver and the organization that manages the system.

Lemma 1.

If there is such that for some , then exists.

Proof.

We prove the contrapositive: If there is no then for any . By definition, we have that  does not exist if and only if and for all

(3)

We must show that this implies . Considering more elements cannot increase the minimum surplus of a set, so from (3), we have for all with . Since is the winner we know that . Joining the inequalities, we have that for any such that , from where we conclude that . ∎

The price in Equation (2) is defined so that it is a threshold for . If is above the threshold then wins and pays that threshold, otherwise does not win and pays nothing. This is called the critical value which is analogous to the second-price in a Vickrey auction. This will be a crucial point in the proof that the mechanism is strategy-proof, which is given in Section 3.2.

3.2 Example

Here we illustrate an instance of the Weighted Minimum Surplus auction. The input graph is presented in Figure 1 and consists of a driver with a start and end position and four passengers. The trips with highlighted paths are in red, in green, in blue and  in yellow with multiple arrows between two nodes indicating an edge being used by multiple highlighted trips. Table 1 shows the input bids, passenger costs and trip costs. We can verify that these four trips are feasible since  for all of , , and . Other feasible trips are omitted because they do not affect the outcome of the auction, so without loss of generality, the set of alternatives is .

The winning trip is since for . Thus, passengers and  are served, and their price must be determined. First, we calculate which depends on  and . Since is the only alternative that does not contain , trivially . By definition  is trip of maximum cardinality such that and . Therefore, and . For , we need do compute and . The second-highest weighted minimum surplus is that of and , therefore . The only trip such that and is itself, therefore . Finally we have . In this example it is interesting to note that even though passenger  had a higher bid than passenger  and both had the same cost, the price was higher.

Passenger
1 14 4 10
2 12 4 8
3 8 4 4
4 10 6 4
(a) Passenger bid, cost and surplus
Trip Passengers
R 5 2 8 16
G 7 3 4 12
B 3 1 10 10
Y 3 1 8 8
(b) Trip cost, size and weighted minimum surplus
Table 1: Inputs to the example
Figure 1: Nodes and edges of example instance. All edges have unit cost.

3.3 Strategy-proofness

In this section, we build theoretical results that are then used in Theorem LABEL:theorem:strategy to show that the WMS auction is strategy-proof.

The objective of a participant in the auction is to maximize their utility. For a served passenger the utility is , the private valuation for the trip minus the price paid . The utility of passenger  if is since and there is no value in not being served. A rational passenger will choose a bid such that the utility is maximized. An auction is strategy-proof when making always maximizes the utility of a passenger.

We see that in the WMS auction the passenger expresses the private valuation in a single value  for winning since the valuation for losing is always zero. So we may say the WMS auction has a single-parameter domain. For discussion on mechanisms with a single-parameter domain see nisan_algorithmic_2007.

The critical value

, given a vector of valuations

by other passengers and an allocation function , is such that for any if then  and if then . An allocation function is allocation monotone if a passenger that wins when reporting a valuation also wins when reporting any higher valuation, considering that and all other bids are held constant. Therefore a passenger cannot go from winning to losing by raising their bid.

To show strategy-proofness, we use the characterization stated in Theorem LABEL:characterization. Proof of this characterization can be found in Theorem 9.36 of nisan_algorithmic_2007.