The Competitive Effects of Variance-based Pricing

Ludwig Dierks, Sven Seuken

Introduction

In infrastructure-heavy industries such as cloud computing, electricity, and mobile data, providers face significant costs associated with maintaining "buffer capacity." This extra capacity is required to handle unpredictable spikes in demand, and its cost is primarily driven by the variance of user demand. Despite this, traditional pricing models typically rely on flat per-unit rates (e.g., dollars per kWh). This creates a "fairness gap" where stable, low-variance users effectively subsidize the higher costs imposed by volatile, high-variance users.

This paper explores the competitive implications of variance-based pricing (VBP), a model where a portion of the price depends on the fluctuations in a user's demand. The authors investigate whether adopting VBP provides a strategic advantage in a competitive market or if providers sticking to traditional flat rates will outcompete innovators.

The Variance-Based Pricing Model

The authors analyze a duopoly serving a continuum of user types, where each user's type $t$ represents their demand variance. Providers compete by setting a linear pricing rule: $$\rho_i(t) = p^f_i + p^v_i t$$ Here, $p^f_i$ is a fixed per-unit price and $p^v_i$ is a variance-based charge. The market consists of conservative providers (who set $p^v = 0$) and innovative providers (who set $p^v \neq 0$). Through a two-period game-theoretic model, the authors solve for the Bayes-Nash Equilibrium (BNE) to determine stable pricing strategies and market outcomes.

Strategic Advantages and Market Power

A key finding is that an innovative provider can leverage the "extra dimension" of variance-based pricing to exert market power. By charging for variance, an innovator can "cherry-pick" the users that are most efficient for them to serve. The authors prove (Theorem 3) that an innovative provider can always choose a pricing strategy that guarantees a higher profit than traditional fixed pricing, regardless of how a conservative competitor responds.

Profit as a function of the fixed price component $p_1^f$, illustrating the best response dynamics.

This strategic flexibility allows providers to escape the "Bertrand Trap"—a scenario where competitors undercut each other's prices until profits vanish. In an asymmetric market where providers have different cost structures (e.g., one is better at handling steady loads while the other excels at bursty loads), VBP allows them to efficiently split the market and increase profits for both parties.

Equilibrium and Social Welfare

The adoption of VBP does more than just shift profits; it improves overall social welfare. By aligning prices with the actual costs users impose on the system, VBP naturally sorts users. Low-variance users migrate to providers that reward stability, while high-variance users are pushed toward providers better equipped to handle volatility.

Profits for both providers at the Bayes-Nash Equilibrium (BNE) relative to the user type cutoff $t$.

In numerical simulations, switching from conservative pricing to a VBP equilibrium improved social welfare (measured as a reduction in total system costs) from -0.2625 to -0.2055. This demonstrates that VBP leads to a more efficient allocation of resources across the infrastructure.

Conclusion and Future Work

Variance-based pricing offers a path toward fairer and more efficient infrastructure markets. While the current study assumes providers can accurately identify user types and that user demand patterns are fixed, it opens several avenues for future research. Key areas include modeling how users might actively smooth their demand to receive better rates and expanding the analysis from a duopoly to an $N$-provider oligopoly. Ultimately, VBP provides a mechanism for providers to stabilize their markets while rewarding users for predictable behavior.

Summary updated

BibTeX
@inproceedings{dierks2020competitive,
  author = {Ludwig Dierks and Sven Seuken},
  title = {{The Competitive Effects of Variance-based Pricing}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2020},
  doi = {10.24963/ijcai.2020/51},
}
Figure