Introduction
Traditional financial models, such as the Black-Scholes formula, rely on rigid assumptions about asset price dynamics, typically assuming Geometric Brownian Motion. However, real-world markets often exhibit "jumps," extreme volatility, and non-normal distributions that violate these assumptions, leading to significant model risk. To address this, researchers seek robust (model-independent) bounds—price ranges that must hold true under the no-arbitrage principle, regardless of the specific path the stock price takes.
Historically, robust bounds have often been too loose to be practically useful or have relied on unrealistic constraints. This paper introduces a novel approach using no-regret learning to derive significantly tighter upper bounds for European and Asian option prices.
Methodology: Hedging as Learning
The core innovation of this work is framing dynamic hedging as an online learning task. An option seller is viewed as a "learner" attempting to minimize "regret"—the difference between their actual portfolio performance and the performance of the best fixed strategy in hindsight.
The $\eta$-Momentum Strategy
The authors propose a new $\eta$-momentum trading strategy for capital allocation between stocks and bonds. This strategy utilizes an exponential potential function where $\eta$ serves as a tuning parameter to optimize the learning rate. By applying Karush-Kuhn-Tucker (KKT) optimization, they determine the minimum initial capital required to ensure the seller can always cover the option's payoff. This minimum capital constitutes the new robust upper bound.
Unlike previous models that required a hard cap on maximum daily returns, this strategy remains effective even when returns are theoretically infinite, making it more suitable for volatile market conditions.
Key Results and Comparative Analysis
The proposed bounds demonstrate superior performance compared to existing model-independent benchmarks across various market scenarios, including different levels of volatility and expiration times.
European Options
For European options, the $\eta$-momentum bounds are notably tighter than previous state-of-the-art results. In simulations, the new bound for at-the-money options was approximately 83% of the bound proposed by DeMarzo et al. (2016).
As shown in the figure above, the improvement is consistent across moneyness levels. While the Black-Scholes price represents an idealized scenario, the robust bounds provide a safe ceiling that accounts for model uncertainty without being excessively conservative.
General Tightness Improvement
The methodology also significantly outperforms the bounds established by Gofer and Mansour (2011). By optimizing the $\eta$ parameter, the authors "squeeze" the price range, providing a much more practical tool for risk management.
Asian Options
A major contribution of this paper is the derivation of the first closed-form robust upper bounds for average-strike Asian options. These options are notoriously difficult to price model-independently because their payoff depends on the average price over the option's life rather than just the terminal price.
Conclusion and Future Directions
By bridging the gap between machine learning and financial engineering, this research provides a practical framework for robust option pricing. The $\eta$-momentum strategy offers tighter, more applicable bounds that do not rely on restrictive assumptions about market dynamics.
While the current work focuses on upper bounds, future research could explore similar no-regret methodologies for deriving robust lower bounds. Additionally, extending this framework to more complex exotic options remains a promising avenue for further study.
Summary updated