Bayesian Signaling and Entry Decisions under Uncertain Market Conditions

Technical Summary: Bayesian Signaling and Entry Decisions under Uncertain Market Conditions

Problem Formulation
This paper addresses the problem of strategic entry deterrence in a continuous-time environment characterized by asymmetric information, irreversible investment, and stochastic market demand. The model features a two-player dynamic Stackelberg game between an incumbent (Firm 1) and a potential entrant (Firm 2). The incumbent possesses private information regarding its "strength" (classified as either strong or weak), while the entrant must infer this type based on observable market outcomes and the incumbent's actions.

The core economic tension involves the incumbent's strategic use of costly advertising and promotional expenditures ($u_1$) to signal strength and deter entry, versus the entrant's optimal stopping problem to determine the timing of market entry ($\tau$) incurring a sunk cost. The market demand ($X$) is modeled not as a standard Geometric Brownian Motion (GBM) or Cox-Ingersoll-Ross (CIR) process, but via the Chan-Karolyi-Longstaff-Sanders (CKLS) stochastic differential equation. This specification allows for mean reversion and, crucially, state-dependent volatility, where the magnitude of uncertainty varies with the level of market demand. The entrant updates its Bayesian beliefs about the incumbent's type continuously, forming a log-likelihood ratio ($Z$) that evolves jointly with the demand process.

Methodology
The authors employ a unified framework integrating stochastic differential games, Bayesian learning, and path-integral control theory.

1. State Dynamics and Belief Updating: The market state $X(s)$ follows a controlled CKLS SDE. The entrant's belief process is reduced to a one-dimensional log-likelihood ratio $Z(s)$, which evolves based on the observed drift difference between strong and weak types. This reduction transforms the infinite-dimensional filtering problem (under a continuous type space) into a tractable Markovian state $(X, Z)$.
2. Equilibrium Characterization: The paper seeks a Markovian Nash feedback equilibrium where firms' expenditure strategies depend on the current state-belief pair. Rather than solving the coupled, nonlinear Hamilton-Jacobi-Bellman (HJB) partial differential equations directly, the authors utilize a Feynman-type path-integral control formulation.
   • This approach involves an exponential transformation of the value function, converting the nonlinear HJB equation into a linear backward equation (resembling a Schrödinger equation).
   • The solution is represented as an expectation over paths of the uncontrolled diffusion process, weighted by an exponential action functional (Feynman-Kac representation).
   • This method bypasses the "curse of dimensionality" associated with grid-based HJB solvers, particularly in high-dimensional state-control spaces, by offering a computational alternative that avoids direct solution of the HJB system.
3. Probabilistic Construction: The paper establishes the rigorous probabilistic foundation for the controlled CKLS process, including the existence of invariant densities, ground-state transformations, and the decomposition of the process into local time and excursion intervals. These results justify the path-integral representation used for equilibrium computation.
4. Numerical Implementation: The equilibrium controls are derived by iteratively solving the coupled nonlinear first-order optimality conditions until convergence. The model is calibrated using quarterly revenue data from Enterprise Products Partners (incumbent) and Targa Resources (entrant) from 2010 to 2024. The revenue series serve as proxies for the market state $X(s)$, while the advertising expenditures are treated as latent equilibrium controls derived from the model.

Key Contributions
The paper makes three primary contributions to the literature on industrial organization and stochastic control:

1. Unified Framework: It integrates CKLS demand uncertainty (with state-dependent volatility), private information, irreversible entry, Bayesian belief updating, and path-integral feedback control into a single continuous-time entry-deterrence model. To the authors' knowledge, this specific combination has not been previously addressed in a unified signaling game.
2. Computational Alternative: It demonstrates the utility of the Feynman-type path-integral approach as a computationally efficient alternative to direct HJB solutions for high-dimensional and nonlinear stochastic control problems. The paper provides a theoretical comparison showing that while HJB methods scale exponentially with dimension ($O(N m^d q^2)$), the path-integral Monte Carlo estimator scales linearly with the number of simulations ($O(M^{-1/2})$) and avoids full grid construction.
3. Empirical Illustration: It provides an empirical application of the theoretical framework using real-world revenue data. The study calibrates the CKLS parameters to reproduce the observed revenue dynamics and generates implied equilibrium expenditure trajectories.

Results

• Theoretical Findings: The analysis characterizes a unique Markovian Nash feedback equilibrium. The results suggest that in markets with persistent shocks, the incumbent's incentive to signal may vanish under certain conditions. Specifically, a weak incumbent may find it optimal to forgo costly signaling when demand is low, effectively revealing its type, while signaling is more prevalent when demand is high but uncertainty is manageable.
• Numerical and Empirical Results:
  • Goodness of Fit: The CKLS specification demonstrates superior fit compared to GBM and CIR benchmarks for the challenger firm (Targa Resources), achieving an $R^2$ of 0.862 versus 0.702 for GBM. For the incumbent, all models perform similarly, though CKLS yields the lowest error metrics.
  • Strategic Trajectories: The simulated equilibrium trajectories reveal distinct strategic behaviors. The incumbent's optimal advertising expenditure follows an increasing, concave path, suggesting a gradual ramp-up of deterrence efforts. Conversely, the entrant's expenditure follows a decreasing, convex path, indicating an initial aggressive entry strategy that tapers off as market conditions and beliefs evolve.
  • Convergence: The trajectories show a convergence in expenditure intensity over time, reflecting a stabilization of the competitive landscape and the resolution of informational asymmetries.
  • Parameter Sensitivity: The calibration reveals a significant asymmetry in control sensitivity, with the entrant's expenditure having a much larger direct impact on the fitted drift of the revenue process than the incumbent's in this specific reduced-form exercise.

Significance and Claims
The paper claims its significance lies in extending existing probabilistic tools (specifically path-integral control) to address a novel industrial organization problem involving dynamic signaling and entry deterrence. The authors explicitly state that the objective is not to introduce a fundamentally new control methodology but to adapt and integrate existing tools to handle the complexities of asymmetric information and state-dependent volatility.

The empirical exercise is presented as an illustrative validation of the model's qualitative implications rather than a formal structural estimation of causal effects. The authors emphasize that the observed revenue data serves as a proxy for the market state, and the resulting expenditure trajectories are model-implied equilibrium controls, not directly observed accounting data. The study concludes that the framework successfully reproduces salient features of strategic behavior—such as persistence, recovery after shocks, and distinct responses based on competitive positions—thereby supporting the theoretical mechanisms of entry deterrence under uncertainty.

The paper acknowledges limitations, including the binary nature of the incumbent's type, the use of reduced-form calibration rather than structural identification, the two-player setting, and the exclusion of jump-diffusion dynamics or regime shifts. These are framed as opportunities for future research rather than flaws in the current approach.