Paper: arXiv 2402.02714
Authors: Changqing Teng, Guanglian Li
Abstract
The rough Bergomi (rBergomi) model can accurately describe the historical and implied volatilities, and has gained much attention in the past few years. However, there are many hidden unknown parameters or even functions in the model. In this work, we investigate the potential of learning the forward variance curve in the rBergomi model using a neural SDE. To construct an efficient solver for the neural SDE, we propose a novel numerical scheme for simulating the volatility process using the modified summation of exponentials. Using the Wasserstein 1-distance to define the loss function, we show that the learned forward variance curve is capable of calibrating the price process of the underlying asset and the price of the European-style options simultaneously. Several numerical tests are provided to demonstrate its performance.
Complexity vs Empirical Score
- Math Complexity: 9.2/10
- Empirical Rigor: 7.1/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper presents advanced stochastic calculus and neural SDE architectures with rigorous mathematical derivations, while also including explicit numerical experiments, performance metrics, and implementation details for the proposed mSOE scheme.
Research Flowchart
flowchart TD
A["Research Goal:<br>Learn Forward Variance Curve in rBergomi"] --> B{"Methodology"}
B --> C["Data: Option Market Prices"]
C --> D["Neural SDE Formulation<br>+ Modified Exponential Summation"]
D --> E["Loss Function:<br>Wasserstein 1-Distance"]
E --> F["Optimization<br>Simultaneous Calibration"]
F --> G["Key Findings"]
G --> H["Simultaneous calibration of<br>Asset Price & European Options"]
G --> I["Efficient solver for<br>Neural SDE volatility process"]
G --> J["Validated via<br>Multiple Numerical Tests"]
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 neural SDE model to learn the forward variance curve of the rough Bergomi model, using a modified sum‑of‑exponentials (mSOE) scheme for simulation and a Wasserstein‑1 loss to calibrate both the underlying asset dynamics and European option prices.
Source sentence
In sum, the contributions of this work are three‑fold. First, we derive an efficient modified sum‑of‑exponentials (mSOE) based method (16) to solve (1)-(2) even with very small Hurst parameter H ∈(0, 1/2). Second, we propose to learn the forward variance curve by the neural SDE using the loss function based on the Wass
Assumes 5
- The forward variance curve ξ0(t) can be any given initial forward variance swap curve consistent with market prices. model
Source sentence
According to [3], ξ0(t) can be any given initial forward variance swap curve consistent with the market price.
- The fractional kernel G(t)=t^{H-1}/2 has a singularity at t=0, which must be handled in simulation. model
Source sentence
The primary difficulty arises from the singularity of the fractional kernel G(t) := t^{H-1}/2
- An equidistant temporal grid is used for discretisation of the SDE. data
Source sentence
Throughout, we consider an equidistant temporal grid 0 = t0 < t1 < · · · < tn = T with a time stepping size τ := T/n and ti := iτ.
- The Hurst parameter H is in (0,1/2), often very small (e.g., H≈0.1). model
Source sentence
First, we derive an efficient modified sum‑of‑exponentials (mSOE) based method (16) to solve (1)-(2) even with very small Hurst parameter H ∈(0, 1/2).
Show 1 more
- Training data are generated by simulating the rBergomi model using the mSOE scheme. other
Source sentence
Vt(θ) in (18) was solved by the mSOE scheme to reduce the computational complexity, then St(θ) was solved using the Euler‑Maruyama method.
Data
- Synthetic rBergomi training samples synthetic unknown
Source sentence
The example code for dataset sampling and network training will be made available at the Github repository https://github.com/evergreen1002/Neural-option-pricing-for-rBergomi-model.
Costs
Not applicable
Finds 4
- The mSOE‑4 scheme already accurately approximates implied volatility curves compared to the exact Cholesky method, showing high efficiency. performance
Source sentence
We observe that mSOE‑4 scheme can already accurately approximate the implied volatility curves given by the exact method. This clearly shows the high efficiency of the proposed mSOE based scheme for simulating the implied volatility curves.
- For N > 15 the mSOE‑N scheme achieves high accuracy in the first and second moments of the Gaussian variable, outperforming the hybrid scheme. performance
Source sentence
We observe that the mSOE‑N based numerical scheme achieves high accuracy for the number of terms N > 15, again clearly showing the high efficiency of the proposed scheme.
- The neural SDE trained with Wasserstein‑1 loss recovers the forward variance curve and matches both the empirical asset distribution and European call prices on the test set. pricing-calibration
Source sentence
It is observed that ST(θ*) and P0(θ*) closely match the exact one on the test set, thereby showing the high accuracy of learning the rBergomi model.
- Training loss and option pricing error can be reduced to an acceptable level within a few hundred iterations, demonstrating feasibility of learning. statistical
Source sentence
In all cases, the training loss can be reduced to an acceptable level, so is the error of pricing, clearly indicating the feasibility of learning within the rBergomi model.
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.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.