pith. sign in

arxiv: 2604.21359 · v1 · submitted 2026-04-23 · 💻 cs.GT

A Markovian Traffic Equilibrium Model for Ride-Hailing

Song Gao , Hanyu Cheng , Chiwei Yan , Guocheng Jiang This is my paper

Pith reviewed 2026-05-08 13:38 UTC · model grok-4.3

classification 💻 cs.GT
keywords ride-hailingtraffic equilibriumsemi-Markov decision processfixed-point iterationcongestion modelingdriver behaviortransportation planningpolicy evaluation
0
0 comments X

The pith

Ride-hailing vehicles reach equilibrium by making sequential order-acceptance and link-choice decisions to maximize total discounted returns in an infinite-horizon semi-Markov decision process.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a Markovian traffic equilibrium model for ride-hailing in which empty and hired vehicles sequentially decide on order acceptance and link choices over a traffic network. These decisions are framed as an infinite-horizon semi-Markov decision process that maximizes each vehicle's total discounted return while endogenizing both competition for passenger demand and link-level congestion. Equilibrium is characterized as the solution to a fixed-point system whose existence is established, and the authors supply relaxed fixed-point iteration algorithms for computing it along with convergence results on specialized networks. Experiments on realistic networks demonstrate the model's applicability to transportation planning and show that omitting congestion or forward-looking behavior produces substantial biases in policy evaluation.

Core claim

We develop a Markovian traffic equilibrium model for ride-hailing in which vehicles, whether empty or hired, make sequential order-acceptance and link-choice decisions over a traffic network to maximize total discounted return in an infinite-horizon semi-Markov decision process. The model endogenizes both competition among empty vehicles for passenger demand and traffic congestion arising from road usage at the link level. We characterize equilibrium as the solution to a fixed-point system, establish its existence, and develop relaxed fixed-point iteration algorithms for equilibrium computation, with convergence results for specialized network structures.

What carries the argument

The fixed-point system that equates each vehicle's optimal policy in the infinite-horizon semi-Markov decision process to the network flows and costs those policies collectively induce.

If this is right

  • Equilibrium flows and policies can be computed via relaxed fixed-point iteration.
  • The iteration algorithms converge on specialized network structures.
  • The model supports transportation planning applications on realistic networks.
  • Policy evaluations that omit congestion effects incur substantial biases.
  • Policy evaluations that omit drivers' forward-looking behavior incur substantial biases.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The framework could be extended to time-varying demand or stochastic passenger arrivals while retaining the fixed-point structure.
  • Planners might apply the model to compare dynamic pricing or matching rules that alter drivers' long-term incentives.
  • Static or myopic models may systematically misstate network flows when drivers plan ahead.
  • Sensitivity analysis on the discount factor could reveal how driver patience affects equilibrium congestion patterns.

Load-bearing premise

Drivers actually follow stationary policies that optimize infinite-horizon discounted returns rather than myopic or finite-horizon rules.

What would settle it

Ride-hailing trajectory and acceptance data from a real city network whose observed link flows and order-acceptance rates deviate systematically from the equilibrium flows predicted by the fixed-point system.

read the original abstract

We develop a Markovian traffic equilibrium model for ride-hailing in which vehicles, whether empty or hired, make sequential order-acceptance and link-choice decisions over a traffic network to maximize total discounted return in an infinite-horizon semi-Markov decision process. The model endogenizes both competition among empty vehicles for passenger demand and traffic congestion arising from road usage at the link level. We characterize equilibrium as the solution to a fixed-point system, establish its existence, and develop relaxed fixed-point iteration algorithms for equilibrium computation, with convergence results for specialized network structures. Computational experiments on realistic networks demonstrate the model's practical value for transportation planning. Ablation analyses reveal that ignoring either traffic congestion or drivers' forward-looking behavior can lead to potentially substantial biases in policy evaluation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The paper develops a Markovian traffic equilibrium model for ride-hailing in which vehicles make sequential order-acceptance and link-choice decisions to maximize total discounted return in an infinite-horizon semi-Markov decision process. Equilibrium is characterized as the solution to a fixed-point system whose existence is established; relaxed fixed-point iteration algorithms are developed for computation, with convergence results stated for specialized network structures. Computational experiments on realistic networks and ablation studies are used to demonstrate practical value for transportation planning and to show that ignoring congestion or forward-looking behavior can induce substantial biases.

Significance. If the fixed-point characterization is valid and the algorithms are reliable on the tested instances, the work provides a useful dynamic game-theoretic framework that jointly endogenizes driver competition for passengers and link-level congestion. The ablation results offer concrete evidence that simpler static or myopic models can produce misleading policy evaluations, which is a strength for applied transportation research.

major comments (1)
  1. [Algorithm development and convergence analysis (cross-referenced with experimental section)] The convergence guarantees for the relaxed fixed-point iteration algorithms are provided only for specialized network structures, yet the computational experiments and ablation studies are conducted on realistic networks. This disconnect means the practical utility claims rest on observed numerical behavior rather than the proven theory, weakening the link between the theoretical development and the reported results.
minor comments (1)
  1. [Abstract] The abstract states existence and convergence results without indicating the proof technique or the precise conditions under which the fixed-point system is well-defined; a brief pointer to the relevant theorem or assumption set would improve readability.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed and constructive report. The single major comment raises a valid point about the scope of our convergence analysis relative to the computational experiments. We address it below and indicate the planned revisions.

read point-by-point responses

  1. Referee: The convergence guarantees for the relaxed fixed-point iteration algorithms are provided only for specialized network structures, yet the computational experiments and ablation studies are conducted on realistic networks. This disconnect means the practical utility claims rest on observed numerical behavior rather than the proven theory, weakening the link between the theoretical development and the reported results.

    Authors: We agree that the theoretical convergence results apply only to specialized network structures (such as acyclic networks or those admitting a contraction-mapping argument under the relaxed iteration). For general networks, including the realistic instances used in the experiments, we rely on observed numerical convergence rather than a general proof. This limitation is already stated in the manuscript, but we acknowledge that the connection between theory and practice could be articulated more clearly to avoid any impression of overclaiming. In the revision we will (i) add an explicit discussion in Section 4.3 (or a new subsection) clarifying the scope of the proven results and the role of empirical evidence, (ii) report additional diagnostics on the observed convergence behavior (iteration counts, sensitivity to the relaxation parameter, and failure cases if any) for the realistic networks, and (iii) temper the language around “practical utility” to emphasize that it is supported by numerical performance on the tested instances. These changes will strengthen the link between the theoretical development and the reported results without requiring new proofs. revision: partial

Circularity Check

0 steps flagged

No circularity in equilibrium fixed-point characterization

full rationale

The derivation begins from an explicit infinite-horizon semi-Markov decision process whose value functions and choice probabilities are defined directly from the driver's discounted-return objective and the network's link-level congestion and demand. Equilibrium is characterized as the fixed point of the resulting system of equations obtained from optimality conditions; this is a standard reduction from the MDP Bellman equations rather than a redefinition or fit. Existence is asserted separately, algorithms are constructed with convergence proven only for special network structures, and experiments on realistic networks are presented as computational validation. No load-bearing self-citations, fitted parameters renamed as predictions, or ansatzes imported via citation appear in the derivation chain. The model is therefore self-contained against its own first-principles assumptions.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the modeling assumption that driver behavior is captured by stationary policies in an infinite-horizon semi-Markov decision process and that equilibrium exists as a fixed point of the resulting flow map. No free parameters or invented entities are mentioned in the abstract.

axioms (2)
  • domain assumption Vehicle decisions can be represented as stationary policies in an infinite-horizon semi-Markov decision process that maximize discounted total return.
    Invoked in the first sentence of the abstract to define the model.
  • domain assumption Equilibrium flows and acceptance rates exist as a solution to a fixed-point system.
    Stated directly when the authors characterize equilibrium.

pith-pipeline@v0.9.0 · 5423 in / 1458 out tokens · 41161 ms · 2026-05-08T13:38:27.303674+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

29 extracted references · 29 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address author booktitle chapter doi edition editor eid howpublished institution isbn issn journal key month note number organization pages publisher school series title type url volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.all := #1...

    work page

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in "" FUNCTION format.date year ...

    work page

  3. [3]

    Transportation Research Part B 30(5):369--386

    Akamatsu T (1996) Cyclic flows, markov process and stochastic traffic assignment. Transportation Research Part B 30(5):369--386

    work page 1996

  4. [4]

    Mathematical Programming 111:33--56

    Baillon JB, Cominetti R (2008) Markovian traffic equilibrium. Mathematical Programming 111:33--56

    work page 2008

  5. [5]

    Transportation Research Part B 129:273--304

    Ban X, Dessouky M, Pang JS, Fan R (2019) A general equilibrium model for transportation systems with e-hailing services and flow congestion. Transportation Research Part B 129:273--304

    work page 2019

  6. [6]

    Journal of Statistical Planning and Inference 21:365--381

    Bhattacharya RN, Majumdar M (1989) Controlled semi-markov models - the discounted case. Journal of Statistical Planning and Inference 21:365--381

    work page 1989

  7. [7]

    Operations Research 67(3):744--769

    Bimpikis K, Candogan O, Saban D (2019) Spatial pricing in ride-sharing networks. Operations Research 67(3):744--769

    work page 2019

  8. [8]

    Bliss L (2019) How much traffic do uber and lyft cause? Urbansim Next ://www.urbanismnext.org/resources/how-much-traffic-do-uber-and-lyft-cause

    work page 2019

  9. [9]

    2017 IEEE 20th International Conference on Intelligent Transportation Systems (ITSC), 1–6 (IEEE Press)

    Calderone D, Shankar S (2017) Infinite-horizon average-cost markov decision process routing games. 2017 IEEE 20th International Conference on Intelligent Transportation Systems (ITSC), 1–6 (IEEE Press)

    work page 2017

  10. [10]

    Business of Apps ://www.businessofapps.com/data/uber-statistics/

    Curry D (2025) Uber revenue and usage statistics (2025). Business of Apps ://www.businessofapps.com/data/uber-statistics/

    work page 2025

  11. [11]

    Transportation research 5(2):83--111

    Dial RB (1971) A probabilistic multipath traffic assignment model which obviates path enumeration. Transportation research 5(2):83--111

    work page 1971

  12. [12]

    Dontchev AL, Rockafellar RT (2014) Implicit Functions and Solution Mappings (Springer)

    work page 2014

  13. [13]

    Howard RA (1971) Dynamic Probabilistic Systems Volume II: Semi-Markov and Decision Processes (John Wiley & Sons, Inc.)

    work page 1971

  14. [14]

    Transportation Research Part C: Emerging Technologies

    Jiang G, Gao S (2026) Joint estimation of a semi-markov model of taxi matching and routing. Transportation Research Part C: Emerging Technologies

    work page 2026

  15. [15]

    Manski C, McFadden D, eds., Structural Analysis of Discrete Data (Cambridge, MA: MIT Press)

    McFadden D (1981) Econometric models of probabilistic choice. Manski C, McFadden D, eds., Structural Analysis of Discrete Data (Cambridge, MA: MIT Press)

    work page 1981

  16. [16]

    Transportation Research Part B 155:135--159

    Oyama Y, Hara Y, Akamatsu T (2022) Markovian traffic equilibrium assignment based on network generalized extreme value model. Transportation Research Part B 155:135--159

    work page 2022

  17. [17]

    Puterman ML (2005) Markov Decision Processes: Discrete Stochastic Dynamic Programming (Wiley-Interscience)

    work page 2005

  18. [18]

    SIAM Journal of Control and Optimization 26(5):1006--1024

    Rust J (1988) Maximum likelihood estimation of discrete control processes. SIAM Journal of Control and Optimization 26(5):1006--1024

    work page 1988

  19. [19]

    Rust J, Traub JF, Wozniakowski H (2002) Is there a curse of dimensionality for contraction fixed points in the worst case? Econometrica 70(1):285--329

    work page 2002

  20. [20]

    The Economist Https://www.economist.com/podcasts/2025/12/02/the-great-wheel-chinas-robotaxi-revolution

    The Economist E (2025) The great wheel: China's robotaxi revolution. The Economist Https://www.economist.com/podcasts/2025/12/02/the-great-wheel-chinas-robotaxi-revolution

    work page 2025

  21. [21]

    Econometrica: Journal of the Econometric Society 29(4):676--699

    Walters AA (1961) The theory and measurement of private and social cost of highway congestion. Econometrica: Journal of the Econometric Society 29(4):676--699

    work page 1961

  22. [22]

    Transportation Research Part B: Methodological 35(9):819--842

    Wong KI, Wong SC, Yang H (2001) Modeling urban taxi services in congested road networks with elastic demand. Transportation Research Part B: Methodological 35(9):819--842

    work page 2001

  23. [23]

    Transportation Research Part B: Methodological 42(10):985--1007

    Wong KI, Wong SC, Yang H, Wu J (2008) Modeling urban taxi services with multiple user classes and vehicle modes. Transportation Research Part B: Methodological 42(10):985--1007

    work page 2008

  24. [24]

    Transportation Science 55(6):1260--1279

    Xu Z, Chen Z, Yin Y, Ye J (2021) Equilibrium analysis of urban traffic networks with ride-sourcing services. Transportation Science 55(6):1260--1279

    work page 2021

  25. [25]

    Naval Research Logistics (NRL) 67(8):705--724

    Yan C, Zhu H, Korolko N, Woodard D (2020) Dynamic pricing and matching in ride-hailing platforms. Naval Research Logistics (NRL) 67(8):705--724

    work page 2020

  26. [26]

    Transportation Research Part B: Methodological 44(8-9):1067--1083

    Yang H, Leung CW, Wong SC, Bell MG (2010) Equilibria of bilateral taxi--customer searching and meeting on networks. Transportation Research Part B: Methodological 44(8-9):1067--1083

    work page 2010

  27. [27]

    Transportation Research Part B 32(4):235--246

    Yang H, Wong SC (1998) A network model of urban taxi services. Transportation Research Part B 32(4):235--246

    work page 1998

  28. [28]

    Transportation Research Part B 177:102819

    Zhang K, Mittal A, Djavadian S, Twumasi-Boakye R, Nie Y (2023) Ride-hail vehicle routing (river) as a congestion game. Transportation Research Part B 177:102819

    work page 2023

  29. [29]

    The International Journal of Robotics Research 35(1-3):186--203

    Zhang R, Pavone M (2016) Control of robotic mobility-on-demand systems: a queueing-theoretical perspective. The International Journal of Robotics Research 35(1-3):186--203

    work page 2016