Paper: arXiv 2306.11061
Authors: Fabio Baschetti, Giacomo Bormetti, Pietro Rossi
Abstract
We propose a neural network-based approach to calibrating stochastic volatility models, which combines the pioneering grid approach by Horvath et al. (2021) with the pointwise two-stage calibration of Bayer et al. (2018) and Liu et al. (2019). Our methodology inherits robustness from the former while not suffering from the need for interpolation/extrapolation techniques, a clear advantage ensured by the pointwise approach. The crucial point to the entire procedure is the generation of implied volatility surfaces on random grids, which one dispenses to the network in the training phase. We support the validity of our calibration technique with several empirical and Monte Carlo experiments for the rough Bergomi and Heston models under a simple but effective parametrization of the forward variance curve. The approach paves the way for valuable applications in financial engineering - for instance, pricing under local stochastic volatility models - and extensions to the fast-growing field of path-dependent volatility models.
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 7.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced neural network architectures and stochastic calculus concepts (e.g., rough volatility models, forward variance curves), indicating high mathematical complexity, while it provides extensive Monte Carlo and empirical experiments with specific model calibrations and mentions implementation details like random grids and two-stage training, demonstrating strong empirical rigor.
Research Flowchart
flowchart TD A["Research Goal: Deep Calibration of Stochastic Volatility Models"] --> B["Data: Generate Implied Volatility Surfaces on Random Grids"] B --> C["Methodology: Combine Random Grid Approach with Pointwise Two-Stage Calibration"] C --> D["Computational Process: Train Neural Network to Learn Calibration Mapping"] D --> E["Outcome: Robust & Accurate Calibration for Rough Bergomi & Heston Models"] E --> F["Applications: Pricing in LSV Models & Path-Dependent Volatility Models"]
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 network-based approach to calibrating stochastic volatility models that combines a grid method with pointwise two‑stage calibration.
Source sentence
We propose a neural network-based approach to calibrating stochastic volatility models, which combines the pioneering grid approach by Horvath et al. (2021) with the pointwise two-stage calibration of Bayer et al. (2018) and Liu et al. (2019).
Assumes 5
- Zero interest rates and dividends are assumed. market
Source sentence
upon assumption of zero interest rates and dividends.
- The spot price follows dSt = St sqrt{Vt}(ρ dBt + sqrt{1-ρ^2} dB⊥t). model
Source sentence
dB⊥t) upon assumption of zero interest rates and dividends.
- The forward variance curve ξ0(t) is piecewise constant. model
Source sentence
The curve is piecewise constant, but its identification precedes the calibration step.
- Parameter sets are drawn uniformly at random from the parameter space. data
Source sentence
A million combinations of rBergomi parameters are drawn uniformly and randomly from the parameter space.
Show 1 more
- Strikes for each maturity are sampled non‑uniformly within a range that mimics market granularity. data
Source sentence
Crucially, strikes are not uniform in [Kmin(T), Kmax(T)] but our sampling tries to replicate similar granularity as the market.
Data
- Synthetic rBergomi training set synthetic unknown
Source sentence
A million combinations of rBergomi parameters are drawn uniformly and randomly from the parameter space. Each is associated with a strike‑expiration pair, and the corresponding option contract is valued using Monte Carlo methods.
- Random‑grid implied volatility surfaces synthetic unknown
Source sentence
We sample a bunch of maturities for each parameter set, and for each maturity, we price a random vector of strikes.
Costs
Not applicable
Finds 4
- The method calibrates the entire market volatility surface in nearly one second without interpolation/extrapolation. pricing-calibration
Source sentence
The result is a small neural network that calibrates the entire market volatility surface in nearly one second and does not require any interpolation/extrapolation.
- Training on random smiles improves out‑of‑sample performance compared to a pure pointwise approach. predictive-accuracy
Source sentence
As we will clearly show, this improves the out‑of‑sample performance of the network (in contrast to the pure pointwise approach) for two reasons.
- A much thinner network achieves comparable accuracy thanks to richer training data. performance
Source sentence
We can work with a much thinner network than Bayer et al. (2018) and Liu et al. (2019) because of the information we include in the training data.
- The approach learns the underlying pricing function extremely well and allows calibration in less than one second on average. pricing-calibration
Source sentence
It learns the underlying pricing function extremely well and it allows for calibration to the market in short time (less than one second on average).
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 34 related papers). Coverage is partial, so this is neither replication nor refutation.
6 papers in the archive
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.