Conjectural Variations in Competitive Dynamic Pricing: A Learning Foundation via Experimentation Design and Feedback Structure
Pith reviewed 2026-05-25 07:18 UTC · model grok-4.3
The pith
Sellers' partial observation of rival prices combined with correlated price experiments causes their learning dynamics to converge to a conjectural variations equilibrium.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
Each seller updates its price using a linear demand estimate based on its own demand data and the competitor prices it observes. The resulting dynamics converge to a conjectural variations equilibrium in which each seller best-responds to a conjecture that rivals' prices co-move systematically with its own price. This conjecture is endogenously determined by the omitted-variable bias that arises when unobserved rivals' prices correlate with the focal seller's experiments. When the bias vanishes, for example under full price feedback or independent experimentation of unobserved rivals, the learning dynamics converge to the standard Nash equilibrium. Simple sufficient conditions on demand are
What carries the argument
The omitted-variable bias in linear demand estimation induced by correlated experimentation under partial price feedback; this bias determines the conjectural variations parameters that govern the long-run equilibrium.
If this is right
- The conjectural variations parameters equal the omitted-variable bias that arises from the correlation structure of experiments and the set of observed prices.
- Full observation of all rival prices removes the bias and produces convergence to Nash equilibrium.
- Independent experimentation by unobserved rivals also removes the bias and produces convergence to Nash equilibrium.
- Under the stated demand conditions the price dynamics converge, and mean squared price error decays at rate ̃O(T^{-1/2}).
Where Pith is reading between the lines
- Market platforms could change long-run competitiveness by altering which prices are shared or by requiring experiment designs that reduce correlation.
- The same bias mechanism may shape outcomes in other multi-agent settings where agents learn payoffs from incomplete observation of others' actions.
- Laboratory experiments that control feedback and experiment correlation could directly test whether the predicted bias appears in subjects' price adjustments.
Load-bearing premise
Demand satisfies conditions that ensure the sellers' price adjustment process converges to an equilibrium.
What would settle it
Simulate the exact learning process with partial price observations and correlated experiments; the long-run price conjectures should equal the omitted-variable bias coefficient computed from the correlation structure, and the mean squared price error should decay at rate roughly T to the minus one half.
read the original abstract
We study competitive dynamic pricing among multiple sellers, motivated by the rise of large-scale experimentation and algorithmic pricing in retail and online marketplaces. Sellers repeatedly set prices using simple learning rules and observe their own realized demand, while possibly observing only a subset of rivals' prices, even though demand depends on all sellers' prices and is subject to random shocks. Each seller runs local price experiments, such as switchback-style designs, and updates a focal price using a linear demand estimate fitted to its own demand data and the competitor prices it observes. Under certain conditions on demand, the resulting dynamics converge to a Conjectural Variations (CV) equilibrium, a classic static equilibrium notion in which each seller best responds under a conjecture that rivals' prices co-move systematically to changes in its own price. Unlike standard CV models that treat conjectures as behavioral primitives, we show that these conjectures arise endogenously from the interaction between the feedback structure and the correlation structure of experimentation. When a seller does not observe some rivals' prices, correlated experimentation induces an omitted-variable bias in demand estimation. We show that this bias determines the conjectures that govern the long-run equilibrium. Notably, when this learning bias vanishes, for example under full price feedback or independent experimentation of unobserved rivals, the learning dynamics converge to the standard Nash equilibrium. We provide simple sufficient conditions on demand for convergence in standard models and establish a finite-sample guarantee, showing that the mean squared price error decays at a rate of $\widetilde O (T^{-1/2})$.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies multi-seller dynamic pricing in which each seller runs local experiments (e.g., switchback designs), fits a linear demand model using its own realized demand and the subset of rival prices it observes, and updates its price accordingly. It claims that, under certain conditions on demand, the resulting stochastic dynamics converge to a Conjectural Variations (CV) equilibrium whose conjectural slopes are exactly the omitted-variable bias terms induced by correlated experimentation when some rival prices are unobserved; when the bias vanishes (full feedback or independent experimentation), the limit is the Nash equilibrium. The paper supplies simple sufficient conditions on demand for standard models and an Õ(T^{-1/2}) finite-sample MSE bound on price error.
Significance. If the identification between omitted-variable bias and CV parameter holds under the stated conditions, the work supplies a statistical micro-foundation for CV equilibria that is directly tied to observable feedback and experimentation structures. The finite-sample rate is a concrete strength that distinguishes the result from purely asymptotic learning analyses in the algorithmic pricing literature.
major comments (2)
- [Abstract / convergence theorem] Abstract and the convergence theorem: the central claim that the fixed point is a CV equilibrium (rather than some other linear-regression fixed point) rests on the assertion that the omitted-variable bias term equals the conjectural slope under the paper's 'certain conditions on demand.' The sufficient conditions are described as simple and applicable to standard models, yet the abstract and the statement of the theorem do not exhibit an explicit verification that this equality survives for common demand specifications containing nonlinear cross-effects or non-constant slopes; without that verification the mapping from bias to CV parameter remains load-bearing and unconfirmed.
- [Finite-sample guarantee] Finite-sample guarantee: the Õ(T^{-1/2}) MSE bound is presented without displayed dependence on the number of sellers, the strength of experimentation correlation, or the dimension of the unobserved price vector. Because these quantities directly affect the omitted-variable bias and the rate at which the bias term stabilizes, the bound's uniformity across multi-seller environments is unclear and affects the practical reach of the convergence result.
Simulated Author's Rebuttal
We thank the referee for the constructive comments, which help clarify the presentation of our results. We address each major comment below.
read point-by-point responses
-
Referee: [Abstract / convergence theorem] Abstract and the convergence theorem: the central claim that the fixed point is a CV equilibrium (rather than some other linear-regression fixed point) rests on the assertion that the omitted-variable bias term equals the conjectural slope under the paper's 'certain conditions on demand.' The sufficient conditions are described as simple and applicable to standard models, yet the abstract and the statement of the theorem do not exhibit an explicit verification that this equality survives for common demand specifications containing nonlinear cross-effects or non-constant slopes; without that verification the mapping from bias to CV parameter remains load-bearing and unconfirmed.
Authors: The sufficient conditions on demand (stated in Theorem 3 and the surrounding discussion) are formulated precisely so that the omitted-variable bias induced by the projection onto observed prices equals the conjectural slope for any demand function whose cross-partials satisfy the stated linearity-in-parameters restriction after conditioning on observables. This covers standard models with nonlinear cross-effects (e.g., linear demand with quadratic interaction terms or constant-elasticity demand after log transformation) because the bias term is defined via the population projection and does not rely on constancy of slopes. We agree, however, that an explicit one-sentence verification for two common specifications would remove any ambiguity; we will add this verification to both the abstract and the theorem statement in the revision. revision: yes
-
Referee: [Finite-sample guarantee] Finite-sample guarantee: the Õ(T^{-1/2}) MSE bound is presented without displayed dependence on the number of sellers, the strength of experimentation correlation, or the dimension of the unobserved price vector. Because these quantities directly affect the omitted-variable bias and the rate at which the bias term stabilizes, the bound's uniformity across multi-seller environments is unclear and affects the practical reach of the convergence result.
Authors: The stated rate treats the number of sellers, the correlation strength, and the dimension of the unobserved price vector as fixed constants; the hidden factors in the Õ notation therefore depend on these quantities through the variance of the design matrix and the size of the omitted-variable bias. The bound is consequently uniform only for fixed market size. We will add an explicit remark after the finite-sample theorem clarifying the dependence of the leading constants on these parameters and noting that the rate remains Õ(T^{-1/2}) whenever the number of sellers grows slower than any positive power of T. revision: partial
Circularity Check
No significant circularity; derivation from bias to CV equilibrium is independent
full rationale
The abstract and description show the conjectures are derived as the omitted-variable bias coefficient in the seller's linear demand regression when some rival prices are unobserved. Convergence to the resulting CV equilibrium is established under separately stated sufficient conditions on demand, with an explicit finite-sample MSE rate. No equations reduce the target equilibrium to a fitted parameter defined by the result itself, no self-citation chain is load-bearing, and the mapping is presented as an endogenous outcome of the feedback/experimentation structure rather than a renaming or self-definition. The result is therefore self-contained against external benchmarks.
Axiom & Free-Parameter Ledger
axioms (1)
- domain assumption Certain conditions on demand allow the dynamics to converge to CV equilibrium
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.