Abstract
We study optimal execution in markets with transient price impact in a competitive setting with $N$ traders. Motivated by prior negative results on the existence of pure Nash equilibria, we consider randomized strategies for the traders and whether allowing such strategies can restore the existence of equilibria. We show that given a randomized strategy, there is a non-randomized strategy with strictly lower expected execution cost, and moreover this de-randomization can be achieved by a simple averaging procedure. As a consequence, Nash equilibria cannot contain randomized strategies, and non-existence of pure equilibria implies non-existence of randomized equilibria. Separately, we also establish uniqueness of equilibria. Both results hold in a general transaction cost model given by a strictly positive definite impact decay kernel and a convex trading cost.
Introduction
In quantitative finance, optimal execution is the process of buying or selling a large block of shares while minimizing "slippage"—the adverse price movement caused by the trader's own activity. When multiple participants trade simultaneously in the same security, the problem becomes an Optimal Execution Game. In these competitive environments, each trader's strategy affects the price for everyone else, creating a complex strategic landscape.
A persistent challenge in this field is that many standard models, such as the widely used Obizhaeva–Wang model, often fail to produce a "pure" Nash equilibrium. In a pure equilibrium, every trader follows a deterministic, fixed plan. In broader game theory, when pure equilibria are absent, researchers typically look to randomized (mixed) strategies to restore balance and find stability. This paper investigates whether allowing traders to use private randomization can resolve the non-existence of equilibria in market impact games.
The De-randomization Principle
The authors analyze a model with $N$ traders facing transient price impact, where the impact of a trade decays over time according to a kernel $G$. They introduce "idiosyncratic randomization," where traders can base their decisions on private random variables (essentially a private random number generator).
The central finding of the research is a "de-randomization" result. The authors prove that for any randomized trading strategy, there exists a deterministic strategy—constructed by taking the "average" of the randomized paths—that results in a strictly lower expected execution cost.
This phenomenon is driven by the mathematical property of strict positive definiteness in the price impact decay kernel. In physical terms, this property ensures that the cost function is strictly convex. Consequently, any "jitter" or variance introduced by randomization acts as a mathematical penalty, increasing the total cost without providing a strategic advantage. By projecting a randomized strategy onto the "market filtration" (the information available to everyone), the authors effectively average out the randomness to find a more efficient path.
Key Results and Uniqueness
The paper establishes two primary theoretical pillars:
- Theorem 4.5 (Suboptimality of Randomization): Any strictly randomized strategy is outperformed by its non-randomized projection onto the market filtration. Because the averaged strategy is always cheaper, a randomized strategy can never be part of a Nash equilibrium.
- Theorem 5.1 (Uniqueness of Equilibrium): The authors extend the framework to show that if a Nash equilibrium exists, it must be unique. This result is robust, applying to models with quadratic costs on trading rates, costs on block trades, and both bounded and singular decay kernels (such as power-law models).
These findings lead to a significant "No-Go" conclusion: if a specific market model lacks a pure Nash equilibrium, allowing for randomized strategies will not fix the problem. The instability is inherent to the model's structure, not the restriction to deterministic strategies. If no pure equilibrium exists, no randomized one exists either.
Methodology
The proofs rely on sophisticated tools from stochastic calculus, specifically predictable and dual predictable projections. These tools allow the authors to mathematically "de-randomize" strategies. The strict positive definiteness of the kernel $G$ acts as a mathematical "penalty" on variance, ensuring that the smoothing of a strategy strictly reduces the expected cost. This approach allows the results to hold across a wide variety of transaction cost models and impact kernels, including those where the initial price impact is theoretically infinite.
Limitations and Future Research
While the paper provides a definitive answer regarding the role of randomization and the uniqueness of equilibria, it does not solve the underlying problem of existence. In many competitive settings, a Nash equilibrium may still fail to exist entirely, and this paper confirms that randomization is not the "fix" researchers hoped for.
Furthermore, the model assumes full information, where traders know the "unaffected" price of the asset. In real-world markets, traders only observe the "affected" price—the sum of the fundamental price and the combined impact of all participants. Future research should focus on "partial information" settings, where traders must infer their competitors' actions from observed price movements, which might lead to different equilibrium behaviors.
Conclusion
This research demonstrates that in the presence of transient price impact, randomization is fundamentally wasteful. By proving that equilibria must be pure and unique, the authors simplify the search for stable trading strategies while highlighting that the fundamental challenges of equilibrium existence in market impact models remain a critical area for future study.
Summary updated