Revenue sharing at music streaming platforms

Basic Information

• Paper ID: 2310.11861
• Title: Revenue sharing at music streaming platforms
• Authors: Gustavo Bergantiños (Universidade de Vigo), Juan D. Moreno-Ternero (Universidad Pablo de Olavide)
• Classification: econ.TH (Economic Theory)
• Publication Date: October 19, 2023
• Paper Link: https://arxiv.org/abs/2310.11861

Abstract

This paper investigates the allocation of subscription revenue from music streaming platforms among content providers (artists). The authors provide axiomatic, game-theoretic, and indirect theoretical foundations for two core methods widely used in practice: pro-rata and user-centric approaches. The pro-rata method allocates revenue proportionally based on artists' total streams; the user-centric method allocates each user's subscription fee proportionally to artists played by that user. The authors also provide theoretical foundations for a family of weighted index methods between these two extremes, addressing the music industry's need to explore new models that better balance the interests of artists, fans, and streaming services.

Research Background and Motivation

Problem Definition

Music streaming platforms (such as Spotify, Apple Music, etc.) generate substantial revenue through subscription fees. Following industry convention, platforms typically allocate approximately 70% of subscription revenue to artists. The core question is: How should this revenue be fairly and reasonably distributed among participating artists based on user listening behavior?

Problem Significance

1. Substantial Economic Scale: The global music streaming market revenue was estimated at $25.84 billion in 2023, with projected user growth to 1.1 billion by 2027
2. Multiple Stakeholders: Involves interests of platforms, artists, record labels, and users
3. Controversial Allocation Methods: The existing pro-rata method suffers from cross-subsidization issues, where high-volume listeners subsidize low-volume listeners, raising fairness concerns
4. Industry Transformation Needs: Platforms like Deezer and SoundCloud have begun transitioning from pro-rata to user-centric methods

Limitations of Existing Methods

1. Pro-rata Method:
   • Considers only total streams, ignoring individual user differences
   • May lead to "superstar phenomenon," with disproportionate income for top artists
   • Creates incentives for "click fraud," where artists may generate false plays
2. User-centric Method:
   • While more equitable, lacks rigorous theoretical foundations
   • Systematic comparison with pro-rata is absent
3. Insufficient Theoretical Research: Existing literature primarily focuses on economic consequences and empirical analysis, lacking normative theoretical foundations

Research Motivation

The authors aim to provide solid normative foundations for revenue allocation methods through multiple theoretical approaches (axiomatic, game-theoretic, and indirect methods), systematically compare the merits and drawbacks of different methods, and provide theoretical guidance for industry practice.

Core Contributions

1. Formal Model Establishment: Proposes a standardized mathematical model of the streaming problem, including artist set, user set, and play matrix
2. Axiomatic Analysis (Theorem 1):
   • Characterizes the family of weighted index methods (satisfying homogeneity and additivity)
   • Proves pro-rata is the unique weighted index satisfying "equal individual impact of similar users"
   • Proves user-centric is the unique weighted index satisfying "equal global impact of users"
3. Game-Theoretic Analysis (Theorems 2-3):
   • Constructs cooperative games, proving they are convex games and completely characterizing the core
   • Proves user-centric allocations always lie in the core, while pro-rata may violate core constraints
   • Further characterizes user-centric through core selection axioms
4. Indirect Method Analysis (Theorem 4):
   • Relates streaming problems to claims problems
   • Proves both methods can be rationalized as two-stage claims rules
5. Normative Evaluation: Introduces and verifies two new axioms—"reasonable lower bound" and "click-fraud prevention"—proving user-centric satisfies them while pro-rata does not
6. Theoretical Support for User-centric: Through multi-faceted analysis, provides strong theoretical support for industry transition from pro-rata to user-centric

Methodology Details

Task Definition

Input: Streaming problem $P = (N, M, t)$, where:

• $N = \{1, ..., n\}$: Set of artists
• $M = \{1, ..., m\}$: Set of users
• $t = (t_{ij})_{i \in N, j \in M}$: Play matrix, where $t_{ij}$ represents the number of times user $j$ plays artist $i$

Assumption: Each user pays a standardized amount of 1 (approximately 70% of actual subscription fee)

Output: Popularity index $I: \mathcal{P} \to \mathbb{R}^N_+$, measuring the importance of each artist

Revenue Allocation: Artist $i$'s revenue is: $R^I_i(N,M,t) = \frac{I_i(N,M,t)}{\sum_{i' \in N} I_{i'}(N,M,t)} \cdot m$

Core Methods

1. Pro-rata Index

The simplest method, allocating by total streams: $P_i(N,M,t) = T_i = \sum_{j \in M} t_{ij}$

Artist $i$'s revenue: $R^P_i(N,M,t) = \frac{T_i}{\sum_{j \in N} T_j} \cdot m$

2. User-centric Index

Each user's subscription fee is allocated proportionally to artists played by that user: $U_i(N,M,t) = \sum_{j \in M} \frac{t_{ij}}{T^j}$

where $T^j = \sum_{i \in N} t_{ij}$ is user $j$'s total plays.

Since $\sum_{i \in N} U_i(N,M,t) = m$, artist $i$'s revenue is directly: $R^U_i(N,M,t) = U_i(N,M,t)$

3. Weighted Index Method Family

A generalized framework where weights depend on users and their play profiles: $I^\omega_i(N,M,t) = \sum_{j \in M} \omega(j, t_{.j}) \cdot t_{ij}$

where $\omega: M \times \mathbb{R}^N_+ \to \mathbb{R}$ is a weighting system, and $t_{.j}$ is user $j$'s play vector.

Special Cases:

• Pro-rata: $\omega(j,x) = 1$ (all users equally weighted)
• User-centric: $\omega(j,x) = \frac{1}{\sum_{i \in N} x_i}$ (weight inversely proportional to total plays)

Technical Innovations

Axiomatic Approach

The authors introduce multiple axioms characterizing the normative properties of different methods:

1. Homogeneity: If each user's plays of artist $i$ are $\lambda$ times those of artist $i'$, the index should maintain this ratio
2. Additivity: If a problem decomposes into sub-problems (e.g., multiple markets), the index should equal the sum of sub-problem indices
3. Equal Individual Impact of Similar Users: For artist $i$, if two users have identical plays of $i$, removing either user should have equal impact on $i$
4. Equal Global Impact of Users: All users should have equal impact on total index (since all pay the same fee)
5. Reasonable Lower Bound: Given user set $C$ and artist set $L_C$ played by them, these artists' revenue should be at least $|C|$
6. Click-Fraud Prevention: When a single user changes play behavior, no artist's revenue change should exceed that user's subscription fee (1)

Game-Theoretic Approach

Construct cooperative game $(N, v^{(N,M,t)})$, where:

• Players: Artist set $N$
• Characteristic function: $v(S) = |\{j \in M: L_j \subseteq S\}|$ (number of users playing only artists in set $S$)

Core Characterization (Theorem 2): $C(N,v) = A(N,M,t)$

where $A(N,M,t)$ is the set of all allocations satisfying: each user's 1 unit is distributed arbitrarily among artists they play, and each artist's revenue is the sum of allocations from all users.

Key Findings:

• This game is convex (supermodular), with large core
• User-centric allocations always lie in the core
• Pro-rata allocations may violate core constraints (Example 1: artist 1 receives 0.2 < 1 = v({1}))

Experimental Setup

Theoretical Verification Method

This is pure theoretical research without traditional experiments, but rather mathematical proofs verifying theoretical results.

Example Analysis

The authors use simple examples (Example 1) throughout to illustrate concepts:

Example 1:

• Users: $M = \{a, b\}$
• Artists: $N = \{1, 2\}$
• Play matrix: $t = \begin{pmatrix} 10 & 0 \\ 0 & 90 \end{pmatrix}$

That is, user $a$ plays only artist 1 (10 times), user $b$ plays only artist 2 (90 times).

Result Comparison:

Intuitive Explanation:

• Pro-rata gives artist 2 much more than artist 1 (due to large difference in total plays)
• User-centric gives both equal revenue (each has one exclusive user)
• User-centric is more reasonable in this scenario

Extended Examples

The authors also consider weighted index examples, adding user $c$ (5 plays of artist 1, 35 plays of artist 2), setting thresholds $\alpha=20, \beta=60$, demonstrating how weighted methods compromise between the two extreme approaches.

Experimental Results

Main Theoretical Results

Theorem 1: Axiomatic Characterization

(a) Weighted Index Family Characterization: An index satisfies homogeneity and additivity if and only if it is a weighted index.

(b) Pro-rata Characterization: An index satisfies homogeneity, additivity, and equal individual impact of similar users if and only if it is the pro-rata index.

(c) User-centric Characterization: An index satisfies homogeneity, additivity, and equal global impact of users if and only if it is the user-centric index.

Significance: Both methods can be viewed as two extremes of the weighted index family—pro-rata "equalizes" plays, user-centric "equalizes" users.

Theorem 2: Core Characterization

For any streaming problem $(N,M,t)$: $C(N,v) = A(N,M,t)$

That is, the core is precisely the set of all allocations distributing each user's 1 unit arbitrarily among artists they play.

Corollaries:

• User-centric allocations always lie in the core (satisfy participation constraints)
• Pro-rata allocations may violate core constraints (violates in Example 1)

Theorem 3: Core Selection Characterization

On restricted domain $\mathcal{P}^*$ (at least 3 users, no user plays all artists), an index satisfies homogeneity, additivity, and core selection if and only if it is the user-centric index.

Significance: Core selection becomes the key distinguishing property of user-centric.

Theorem 4: Claims Problem Connection

(a) Both pro-rata and user-centric can be expressed as weighted proportional rules

(b) $R^P(N,M,t) = R^{P,P}(N,K,c,E)$ (two-stage proportional rule)

(c) $U(N,M,t) = R^{CEA,P}(N,K,c,E)$ (constrained equal awards + proportional rule)

Axiom Verification Results

Key Findings: User-centric outperforms pro-rata on multiple normative axioms, particularly:

1. Satisfies reasonable lower bound (user fees should flow to artists they play)
2. Prevents click fraud (single user manipulation is limited)
3. Guarantees core allocation (artists have no exit incentive)

Case Analysis

Deep Analysis of Example 1:

• Take $S = \{1\}$, then $v(S) = 1$ (user $a$ plays only artist 1)
• Pro-rata gives artist 1 revenue $R^P_1 = 0.2 < 1 = v(S)$, violating core constraint
• Artist 1 has incentive to leave platform and create new platform (can take user $a$ and their 1 unit fee)
• User-centric gives artist 1 revenue $R^U_1 = 1$, satisfying core constraint

Practical Significance: Pro-rata may cause artist attrition, threatening platform stability.

Axiom Independence

The authors prove through counterexamples that all axioms in the characterizations are independent (cannot be omitted), for example:

• $I^1_i = 1$ (uniform index) satisfies additivity but not homogeneity
• $I^2_i = \sum_j (t_{ij} + \frac{T_i}{T^j + \sum_{i'} T_{i'}})$ satisfies homogeneity but not additivity

Related Work

Most Relevant Research

Alaei et al. (2022):

• Also studies pro-rata vs user-centric
• Focus: When can artist participation be maintained, platform and artist preferences
• Main finding: Artists with high-volume listeners prefer pro-rata
• This paper's distinction: Provides normative theoretical foundations rather than positive economic analysis

Empirical Research

1. Haampland et al. (2022): Using French platform data (427 million plays), finds user-centric:
   • Better aligns consumer choice with revenue distribution
   • Reduces superstar phenomenon
   • Favors organic plays over curated plays
2. Meyn et al. (2023): Using Spotify data, estimates transition to user-centric would cause approximately €170 million/year redistribution, mainly from mainstream to niche genres
3. Muikku (2017): The lower the total plays, the larger the revenue difference between methods

Theoretical Literature

Axiomatic Resource Allocation

• Thomson (2023): Survey of axiomatic economic design
• Bergantiños & Moreno-Ternero (2020): Sports broadcasting revenue allocation (similar structure)
• Singal et al. (2022): Online advertising attribution problem

Cooperative Game Theory

• Harsanyi (1959): Dividend decomposition
• Vasil'ev (1978, 1981): Harsanyi set and core relationship
• van den Brink et al. (2014): Constrained core solutions

Claims Problems

• O'Neill (1982): Foundational work on bankruptcy problems
• Ju et al. (2007): Multi-issue claims problems
• Thomson (2019): Survey of claims problems

Industrial Organization Literature

• Adams & Yellen (1976): Commodity bundling theory
• Aguiar & Waldfogel (2018, 2021): Streaming impact on music industry
• Rietveld & Schilling (2020): Platform competition survey

Conclusions and Discussion

Main Conclusions

1. Theoretical Unification: Provides unified theoretical framework for revenue allocation through axiomatic, game-theoretic, and claims problem approaches
2. Method Comparison:
   • Both pro-rata and user-centric have solid theoretical foundations but different emphases
   • Pro-rata emphasizes play equality, user-centric emphasizes user equality
   • User-centric outperforms pro-rata on multiple normative dimensions
3. Practical Recommendations:
   • Technological advances have eliminated user-centric implementation barriers
   • User-centric better aligns with intuition that "user fees should flow to artists they choose"
   • Satisfies core constraints, preventing artist attrition
4. Flexible Solutions: Weighted index family provides intermediate choices, adjustable to platform characteristics

Limitations

1. Model Simplifications:
   • Assumes all users pay identical fees (reality has different subscription tiers)
   • Ignores other revenue sources like advertising
   • Play unit definition (30-second rule) is controversial
2. Cooperative Game Assumptions:
   • Characteristic function $v(S)$ uses pessimistic estimate (only exclusive users)
   • Some users (e.g., 99% plays outside coalition artists) might also leave in reality
3. Theoretical Scope:
   • Primarily focuses on music streaming; applicability to Netflix-like general content platforms needs further research
   • Ignores dynamic incentives (artist creation quality, platform curation, etc.)
4. Computational Complexity: Does not discuss computational efficiency for large-scale platform implementation

Future Directions

1. Dynamic Models: Consider multi-period problems, artist creation incentives, and user behavior evolution
2. Heterogeneous Users: Differential treatment of different subscription tiers and ad-supported users
3. Platform Competition: Revenue allocation strategies in multi-platform environments
4. Empirical Validation: Verify theoretical predictions using real platform data
5. Extended Applications: Apply framework to podcasts, video streaming, and other content platforms
6. Fairness Deepening: Further evaluate methods incorporating fairness theory (e.g., Rawlsian justice)

In-Depth Evaluation

Strengths

1. Theoretical Rigor:
   • Provides complete axiomatic characterizations with proven axiom independence
   • Game-theoretic analysis completely characterizes core with clear conclusions
   • Mathematical proofs are rigorous (detailed in appendix)
2. Method Diversity:
   • Three independent approaches (axiomatic, game-theoretic, claims problem) mutually validate
   • Multi-faceted analysis enhances conclusion credibility
3. Practical Relevance:
   • Directly addresses industry disputes (Deezer, SoundCloud platform transitions)
   • Provides actionable policy recommendations
   • Examples are concise and powerfully illustrate method differences
4. Literature Integration:
   • Comprehensively reviews literature across economics, game theory, resource allocation
   • Clearly positions paper's contributions
5. Normative Insights:
   • Reasonable lower bound and click-fraud prevention axioms have strong intuitive appeal
   • Core selection connection to artist participation incentives has practical significance

Weaknesses

1. Model Limitations:
   • Static model, ignores temporal dimension and dynamic incentives
   • Overlooks platform curation, recommendation algorithms, and other practical complexities
   • Assumes all plays equally valuable (30-second vs. full play)
2. Empirical Absence:
   • Pure theory without real data validation
   • No quantification of actual revenue distribution differences between methods
   • Lacks empirical support for artist behavior responses (creation incentives)
3. Weighted Method Family:
   • While introducing weighted index family, lacks guidance on specific weight system selection
   • Example's piecewise weight function is somewhat ad hoc
   • Insufficient discussion on practical weight selection
4. Core Game Limitations:
   • Characteristic function $v(S)$ definition may be overly conservative
   • Ignores intermediate user loss scenarios
   • Insufficient discussion of realistic feasibility of artist coalition formation
5. Cross-Platform Issues:
   • Ignores artists releasing on multiple platforms
   • Unaddressed impact of inter-platform competition on revenue allocation

Impact

1. Academic Contribution:
   • Exemplifies application of classical microeconomic theory tools to emerging digital economy problems
   • May inspire similar research on other platform economy revenue allocation (short video, podcasts, etc.)
   • Methodological integration of multiple theoretical approaches provides valuable paradigm
2. Industry Impact:
   • Provides theoretical support for ongoing industry transformation
   • May influence regulatory policy (e.g., EU streaming platform regulation)
   • Helps platforms design fairer revenue allocation mechanisms
3. Policy Significance:
   • Supports creator rights protection
   • Promotes healthy digital content market development
   • Provides normative framework for platform governance
4. Reproducibility:
   • Theoretical results fully verifiable (mathematical proofs)
   • Clear concept definitions, easy to understand and apply
   • Practical implementation requires platform data support

Applicable Scenarios

1. Direct Application:
   • Music streaming platform revenue allocation policy design
   • Platform communication tool explaining allocation mechanisms to users
   • Regulatory agency fairness assessment
2. Analogous Applications:
   • Podcast platforms (e.g., Spotify Podcasts)
   • Audiobook platforms (e.g., Audible)
   • Short video creator revenue sharing (YouTube, TikTok)
   • Online education platform course revenue allocation
3. Theoretical Extensions:
   • Multi-sided platform market design
   • Sharing economy revenue allocation (Uber, Airbnb)
   • Open-source software contributor incentives
4. Inapplicable Scenarios:
   • Per-play payment models (no subscription fee pooling)
   • Highly differentiated pricing content markets
   • Scenarios requiring subjective content quality evaluation

Selected References

1. Alaei et al. (2022): Revenue-Sharing Allocation Strategies for Two-Sided Media Platforms. Management Science 68, 8699-8721.
2. Haampland et al. (2022): Fairness and royalty payment systems on music streaming platforms. Mimeo.
3. Meyn et al. (2023): Consequences of platforms' remuneration models for digital content. Journal of the Academy of Marketing Science 51, 114-131.
4. Thomson (2019): How to divide when there isn't enough. Econometric Society Monograph.
5. Bergantiños & Moreno-Ternero (2020): Sharing the revenues from broadcasting sport events. Management Science 66, 2417-2431.

Overall Assessment: This is an excellent economic theory paper with rigorous theory and strong practical relevance. The authors skillfully apply classical tools—axiomatic methods, cooperative game theory, and claims problems—to novel digital platform economy questions, providing solid normative theoretical foundations for industry disputes. While the model involves certain simplifications, the core insight—that user-centric methods outperform pro-rata on multiple normative dimensions—carries important theoretical and practical value. The paper provides strong theoretical support for music streaming revenue allocation reform and establishes a methodological paradigm for related platform economy research.