Paper: arXiv 2401.08600

Authors: Bernhard Hientzsch

Abstract

We consider two data driven approaches, Reinforcement Learning (RL) and Deep Trajectory-based Stochastic Optimal Control (DTSOC) for hedging a European call option without and with transaction cost according to a quadratic hedging P&L objective at maturity (“variance-optimal hedging” or “final quadratic hedging”). We study the performance of the two approaches under various market environments (modeled via the Black-Scholes and/or the log-normal SABR model) to understand their advantages and limitations. Without transaction costs and in the Black-Scholes model, both approaches match the performance of the variance-optimal Delta hedge. In the log-normal SABR model without transaction costs, they match the performance of the variance-optimal Barlett’s Delta hedge. Agents trained on Black-Scholes trajectories with matching initial volatility but used on SABR trajectories match the performance of Bartlett’s Delta hedge in average cost, but show substantially wider variance. To apply RL approaches to these problems, P&L at maturity is written as sum of step-wise contributions and variants of RL algorithms are implemented and used that minimize expectation of second moments of such sums.

Complexity vs Empirical Score

  • Math Complexity: 8.5/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced mathematical constructs like stochastic optimal control and quadratic hedging P&L objectives, requiring dense derivations and heavy LaTeX formulas. It is empirically rigorous with data-driven methods, simulating hedging under Black-Scholes and SABR models with transaction costs, and includes comparative performance analysis of RL and DTSOC agents, though it lacks real-market backtests.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Final Quadratic Hedging<br>with/without Transaction Costs"] --> B{"Methodology"}
  
  B --> C["Reinforcement Learning<br>RL Minimizing Expectation<br>of Second Moments"]
  B --> D["Deep Trajectory-based Stochastic<br>Optimal Control DTSOC"]
  
  C --> E["Market Environments"]
  D --> E
  
  E --> F["Black Scholes Model"]
  E --> G["Log Normal SABR Model"]
  
  F --> H["Key Findings"]
  G --> H
  
  H --> I["Without Transaction Costs<br>Match Variance Optimal Delta Hedge<br>in Black Scholes"]
  H --> J["Without Transaction Costs<br>Match Bartlett s Delta Hedge<br>in SABR Model"]
  H --> K["Trained on BS / Tested on SABR<br>Match Average Cost of Bartlett<br>But Wider Variance"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes two data‑driven approaches, Reinforcement Learning (RL) and Deep Trajectory‑based Stochastic Optimal Control (DTSOC), for hedging a European call option under a quadratic P&L objective.

Source sentence
We consider two data driven approaches, Reinforcement Learning (RL) and Deep Trajectory-based Stochastic Optimal Control (DTSOC) for hedging a European call option without and with transaction cost according to a quadratic hedging P&L objective at maturity ("variance-optimal hedging" or "final quadratic hedging").

Assumes 5

  • The hedging instruments are traded liquidly and hedging does not impact their prices. market
    Source sentence
    We assume here that the hedging instruments are traded liquidly and that the hedged instrument is such that it can be hedged without impacting the prices of the hedging instruments.
  • The stochastic volatility in the log‑normal SABR test case is treated as a latent, unobserved state. model
    Source sentence
    We treat it here as a latent state that is not provided as input to the agent/policy.
  • Stock dynamics follow either Black‑Scholes with constant volatility or log‑normal SABR with stochastic volatility. model
    Source sentence
    We assume the Black‑Scholes or SABR model for the simulation environment where the stock dynamics is given by dSt = µSt dt + σtSt dWt
  • Zero interest rates are assumed for the analytical variance‑optimal hedge derivations. statistical
    Source sentence
    for the case of zero interest rates) and applies them to the log‑normal SABR and the Rough Bergomi model.
Show 1 more
  • P&L at maturity can be written as a sum of step‑wise contributions for RL optimization. other
    Source sentence
    To apply RL approaches to these problems, P&L at maturity is written as sum of step‑wise contributions and variants of RL algorithms are implemented and used that minimize expectation of second moments of such sums.

Data

  • Synthetic Black‑Scholes trajectories synthetic unknown
    Source sentence
    We assume the Black‑Scholes or SABR model for the simulation environment where the stock dynamics is given by dSt = µSt dt + σtSt dWt
  • Synthetic log‑normal SABR trajectories synthetic unknown
    Source sentence
    We assume the Black‑Scholes or SABR model for the simulation environment where the stock dynamics is given by dSt = µSt dt + σtSt dWt

Costs

Modeled Costs are part of the reported headline results.

Finds 5

  • Without transaction costs in the Black‑Scholes model, both RL and DTSOC match the performance of the variance‑optimal Delta hedge. performance
    Source sentence
    Without transaction costs and in the Black‑Scholes model, both approaches match the performance of the variance‑optimal Delta hedge.
  • Without transaction costs in the log‑normal SABR model, both approaches match the performance of the variance‑optimal Bartlett’s Delta hedge. performance
    Source sentence
    In the log‑normal SABR model without transaction costs, they match the performance of the variance‑optimal Barlett’s Delta hedge.
  • For non‑zero and increasing transaction costs, both RL‑QH and deep‑QH outperform the variance‑optimal Delta hedge on average, with deep‑QH doing so more consistently in the Black‑Scholes case. performance
    Source sentence
    For non‑zero and increasing transaction costs, both RL‑QH and deep‑QH outperform the variance optimal Delta hedging on average, with deep‑QH doing so more consistently than RL‑QH in the Black‑Scholes case.
  • Agents trained on Black‑Scholes but tested on SABR trajectories match Bartlett’s Delta hedge in average cost, but exhibit larger variance (deep‑QH) and dramatically larger variance (RL‑QH). performance
    Source sentence
    On average, RL‑QH and deep‑QH still match the performance of Bartlett’s Delta hedge, however showing larger variance (for deep‑QH) and dramatically larger variance (RL‑QH).
  • Deep‑QH and RL‑QH agents match the (approximate) variance‑optimal strategies in average cost, with RL‑QH showing a wider spread of extreme gains and losses. performance
    Source sentence
    Deep‑QH and RL‑QH agents match the (approximate) variance optimal strategies in average cost with comparable (deep‑QH) or wider (RL‑QH) variance and range, with RL‑QH allowing a wider range of both extreme gains and losses.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 26 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 1 paper in the archive (1 by independent authors): A Topological Approach to Parameterizing Deep Hedging … (uses its method or data)

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.