Paper: arXiv 2510.10343
Authors: Giorgia Rensi, Pietro Rossi, Marco Bianchetti
Abstract
The SABR model is a cornerstone of interest rate volatility modeling, but its practical application relies heavily on the analytical approximation by Hagan et al., whose accuracy deteriorates for high volatility, long maturities, and out-of-the-money options, admitting arbitrage. While machine learning approaches have been proposed to overcome these limitations, they have often been limited by simplified SABR dynamics or a lack of systematic validation against the full spectrum of market conditions. We develop a novel SABR DNN, a specialized Artificial Deep Neural Network (DNN) architecture that learns the true SABR stochastic dynamics using an unprecedented large training dataset (more than 200 million points) of interest rate Cap/Floor volatility surfaces, including very long maturities (30Y) and extreme strikes consistently with market quotations. Our dataset is obtained via high-precision unbiased Monte Carlo simulation of a special scaled shifted-SABR stochastic dynamics, which allows dimensional reduction without any loss of generality. Our SABR DNN provides arbitrage-free calibration of real market volatility surfaces and Cap/Floor prices for any maturity and strike with negligible computational effort and without retraining across business dates. Our results fully address the gaps in the previous machine learning SABR literature in a systematic and self-consistent way, and can be extended to cover any interest rate European options in different rate tenors and currencies, thus establishing a comprehensive functional SABR framework that can be adopted for daily trading and risk management activities.
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 8.5/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced stochastic calculus and Monte Carlo simulation for the SABR model, reflecting high mathematical complexity, while demonstrating robust empirical rigor through massive synthetic datasets and systematic validation against real market volatility surfaces.
Research Flowchart
flowchart TD
Goal["Research Goal: Develop an accurate<br>arbitrage-free SABR calibration tool<br>addressing Hagan's limitations"]
Data["('Training Data<br>200M+ Points<br>MC Simulation of SABR Dynamics<br>(30Y maturity, extreme strikes)')"]
Method["Model: SABR DNN Architecture<br>Specialized Deep Neural Network"]
Train["Training Process<br>Learn True SABR Stochastic Dynamics"]
Calib["Arbitrage-Free Calibration<br>(Real Market Surfaces<br>Negligible Computational Effort)"]
Outcome["Outcome: Comprehensive Functional SABR Framework<br>- Daily Trading & Risk Management<br>- Extensible to any IR European Options<br>- No Retraining Required"]
Goal --> Data
Data --> Method
Method --> Train
Train --> Calib
Calib --> Outcome
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 novel SABR DNN, a specialized deep neural network architecture that learns the true SABR stochastic dynamics and provides arbitrage‑free calibration and pricing of Cap/Floor volatility surfaces.
Source sentence
We develop a novel SABR DNN, a specialized Artificial Deep Neural Network (DNN) architecture that learns the true SABR stochastic dynamics using an unprecedented large training dataset of interest rate Cap/Floor volatility surfaces...
Assumes 6
- The shift parameter is fixed to λ = 3% consistent with EUR Cap/Floor market quotations. model
Source sentence
The shift parameter was fixed to λ = 3%, consistently with market quotations for EUR Caps/Floors which include negative strikes (see app. A.2).
- The β parameter is not fixed and is calibrated together with other SABR parameters, preserving full model flexibility. model
Source sentence
In particular, the β parameter is not fixed (as done e.g. in [31]), but it is calibrated together with the other model parameters, preserving the full flexibility of the original SABR formulation.
- A simple log‑Euler discretization scheme is used for the SABR dynamics. model
Source sentence
In this paper we adopt a simple log‑Euler discretization scheme of the SABR model dynamics.
- Input features are standardized to zero mean and unit variance using StandardScaler. data
Source sentence
The DNNs’ input data are standardized using the StandardScaler from scikit-learn, which transforms each input feature to have zero mean and unit variance...
Show 2 more
- Monte Carlo simulations are high‑precision and unbiased, providing the ground‑truth volatilities. model
Source sentence
Our dataset is obtained via high‑precision unbiased Monte Carlo simulation of a scaled shifted‑SABR stochastic dynamics...
- The training dataset covers very long maturities (up to 30Y) and extreme strikes consistent with market quotations. data
Source sentence
including very long maturities (30Y) and extreme strikes consistently with market quotations.
Data
- Synthetic SABR Cap/Floor volatility surface dataset synthetic unknown
Source sentence
more than 200 million points of interest rate Cap/Floor volatility surfaces, including very long maturities (30Y) and extreme strikes...
- EUR Cap/Floor market quotations options unknown
Source sentence
with market quotations for EUR Caps/Floors which include negative strikes
Costs
Not applicable
Finds 5
- The three DNNs achieve RMSE around 0.26%–0.28% on training and test sets, comparable to the typical EUR Cap/Floor bid‑ask spread (~0.25%). performance
Source sentence
We observe in tab. 6 that both training and test performances in terms of RMSE are quite good with respect to the typical volatility bid‑ask spread observed on the EUR Cap/Floor market (≈0.25% for the most liquid options).
- The DNN volatility approximation error is largest for short maturities and extreme strikes, but remains within Monte Carlo errors. pricing-calibration
Source sentence
The largest discrepancies are associated with short maturities and extreme strikes, where the MC SABR volatilities show the largest MC errors
- The DNN acts as a filter, extracting true information while discarding Monte Carlo noise. theoretical
Source sentence
we observe that the DNNs act as filters, extracting true information while discarding noise, which is reflected in the halo observed around the noisier points.
- Once trained, the SABR DNN can calibrate online real‑market volatility surfaces on any business date with very high precision and negligible computational effort. pricing-calibration
Source sentence
Once trained, our SABR DNN is able to calibrate online the SABR model parameters to real market volatility surfaces quoted on different business dates with very high precision and computational performance.
- The DNN can price Cap/Floor instruments for any IBOR tenor without needing the tenor as an input. performance
Source sentence
our SABR DNN “knows” only the forward rate fixing date, not the EURIBOR tenor nor the Caplet/Floorlet maturity date. Hence, once trained, it is able to calibrate Caplet/Floorlet volatilities on any IBOR tenor.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 37 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.