Paper: arXiv 2312.14044
Authors: Roberto Daluiso, Marco Pinciroli, Michele Trapletti, Edoardo Vittori
Abstract
This work studies the dynamic risk management of the risk-neutral value of the potential credit losses on a portfolio of derivatives. Sensitivities-based hedging of such liability is sub-optimal because of bid-ask costs, pricing models which cannot be completely realistic, and a discontinuity at default time. We leverage recent advances on risk-averse Reinforcement Learning developed specifically for option hedging with an ad hoc practice-aligned objective function aware of pathwise volatility, generalizing them to stochastic horizons. We formalize accurately the evolution of the hedger’s portfolio stressing such aspects. We showcase the efficacy of our approach by a numerical study for a portfolio composed of a single FX forward contract.
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 6.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper introduces advanced mathematical concepts like stochastic horizons and risk-averse RL with detailed formalization, indicating high math complexity; the empirical section includes a numerical study on an FX forward with transaction costs and collateral, showing implementation-heavy backtesting, but without public code/datasets, leading to a mid-high rigor score.
Research Flowchart
flowchart TD A["Research Goal: CVA Hedging<br>for FX Forwards"] --> B["<b>Problem:</b><br>Sensitivities-based hedging<br>sub-optimal due to costs<br>& realistic constraints"] B --> C["<b>Methodology:</b><br>Risk-Averse Reinforcement Learning<br>Stochastic Horizon Policy<br>Pathwise Volatility Objective"] C --> D["<b>Inputs:</b><br>FX Forward Portfolio<br>Market Data (Spot, Vol, Rates)<br>Bid-Ask Spreads"] D --> E["<b>Process:</b><br>RL Agent optimizes<br>hedging strategy over<br>stochastic time paths"] E --> F["<b>Key Findings:</b><br>RL outperforms delta<br>hedging; effectively<br>manages CVA & costs"]
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 a risk‑averse reinforcement‑learning framework with stochastic horizons for CVA hedging, including a Trust‑Region Volatility Optimization generalization and an importance‑sampling scheme.
Source sentence
We leverage recent advances on risk‑averse Reinforcement Learning developed specifically for option hedging with an ad hoc practice‑aligned objective function aware of pathwise volatility, generalizing them to stochastic horizons.
Assumes 6
- Foreign currencies are always exchanged against the reference currency and never between themselves. execution
Source sentence
In the above modelling, we are supposing that foreign currencies are always exchanged against the reference currency and never between themselves.
- Interest rates are known time‑independent constants for all cash accounts. market
Source sentence
We also assume that interest rates are known time‑independent constants for all cash accounts.
- FX rate dynamics follow a Geometric Brownian Motion with constant drift µP,φ and volatility σP,φ. model
Source sentence
We simulate the FX rate via a Geometric Brownian Motion (GBM) dφt/φt = µP,φ dt + σP,φ dW P,φ t , with µP,φ the drift, σP,φ the volatility.
- Default intensity follows a Cox‑Ingersoll‑Ross (CIR) process with constant parameters. model
Source sentence
We simulate the default intensity via the Cox Ingersoll Ross (CIR) model
Show 2 more
- Real‑world correlation between FX and default intensity drivers is zero. statistical
Source sentence
Real‑world correlation of Brownian drivers ρP λφ 0%.
- CDS trading cost is captured by a constant parameter γλ. execution
Source sentence
CDS trading cost parameter γλ 1.66 × 10⁻³
Data
- Synthetic FX and default intensity data synthetic unknown
Source sentence
... We simulate the default intensity via the Cox Ingersoll Ross (CIR) model
Costs
Modeled
Transaction costs are modeled as Tt(N)=∑h∈H ¯T h t, with bid‑ask spreads φc¯c,± and cost parameters γλ for CDS.
Source sentence
Finally, we can compute the total transaction costs as Tt(N) :=
Finds 3
- Importance sampling reduces noise in training and improves the risk‑averse objective ηπ compared to naive sampling. performance
Source sentence
The noise reduction effect of importance sampling in training is confirmed whichever the risk aversion ...
- Optimized agents consistently save on hedging costs by reducing hedge rebalances and exploiting differences between risk‑neutral and real‑world probabilities. performance
Source sentence
The optimized agents are always able to save on costs by a wise reduction of hedge rebalances, but also cleverly exploit the expected returns arising from any difference between risk‑neutral and real‑world probabilities...
- Agents trained with the proposed method outperform benchmark delta‑hedge and jump‑hedge strategies across a range of risk‑aversion parameters. performance
Source sentence
Even without importance sampling, we do get agents with an interesting risk‑cost profile when compared to the two benchmarks
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 17 related papers). Coverage is partial, so this is neither replication nor refutation.
1 paper in the archive
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.