ICAIF '26 Tutorial

Amortized Bayesian Calibration of Volatility Models

When
Sunday • 15 November 2026
11:00–13:00 CET
Where
Room 3 • Bocconi University
Via Röntgen 1 • Milan

View Program View Materials GitHub Repository

Overview

This tutorial introduces amortized Bayesian inference for volatility models using BayesFlow. We will train neural networks to rapidly calibrate volatility models to financial data.

The examples will demonstrate the fundamental principles of amortized Bayesian workflows using popular volatility models, such as Heston and rough Bergomi.

Illustrative volatility-smile curves with a narrow spread at quoted strikes and a wider spread beyond them.

Learning outcomes

  • Formulate volatility-model calibration as Bayesian inference, specifying a prior and an observation model.
  • Use BayesFlow to train a neural posterior estimator on simulated data and apply it to new observations.
  • Assess posterior approximations using simulation-based calibration and posterior predictive checks.

Presenters

Francesco Muia

Francesco Muia

Muia AI & Analytics SolutionsPyMC Labs

AI and Bayesian data science consultant; PhD in Theoretical Physics and Executive MBA. Extensive research and industry experience in computational finance.

Alexander Fengler

Alexander Fengler

Brown University

Postdoctoral Fellow; PhD in Computational Cognitive Science and MPhil in Statistics. Extensive research experience in computational finance, open-source probabilistic modeling, and SBI.

Stefan T. Radev

Stefan T. Radev

Rensselaer Polytechnic Institute

Assistant Professor and lead developer of BayesFlow. PhD in Statistical Modeling. AI researcher and Bayesian statistician with extensive experience in teaching, workshops, and consulting.

Jerry M. Huang

Jerry M. Huang

Rensselaer Polytechnic Institute

Postdoctoral Scholar in Statistical Modeling and core BayesFlow developer. Specializes in computational modeling, simulation-based inference, and digital twins across different disciplines.

Program

Sunday 15 November, 11:00–13:00 CET.

Time Session Presenters Materials
11:00-11:15 Introduction: Bayes for volatility models Francesco Muia Slides
11:15-11:30 Amortized Bayesian inference Alexander Fengler TBD
11:30-11:45 Neural network architectures Stefan T. Radev TBD
11:45-12:30 Worked examples: Heston, rough Bergomi and return histories Alexander Fengler
Francesco Muia
Heston and rough Bergomi
Return histories
12:30-13:00 Advanced Topics and Open Question Stefan T. Radev
Jerry M. Huang
To be added

Materials

Session notebooks are linked in the program.

Notebooks

Application notebooks in marimo. A laptop CPU is enough.

Code

The tutorial repository, with the simulators, the trained networks and setup instructions.

Before the tutorial

Bring a laptop and install the tutorial environment before the session. Pixi installs Python 3.12, BayesFlow 2.1, marimo and their dependencies. A CPU is enough to work through the saved examples.

macOS (Apple Silicon)

The supplied macOS environment is for Apple Silicon (M1 or later). Intel Macs are not covered by the lockfile.

  1. Install Pixi. Open Terminal and run:
    curl -fsSL https://pixi.sh/install.sh | sh

    Close and reopen your terminal, then check that pixi --version works.

  2. Download the tutorial. Download the repository ZIP and extract it into your Downloads folder. You should have a folder named sbi4finance-main.
  3. Install, check and open a notebook. In the new terminal, run:
    cd ~/Downloads/sbi4finance-main/tutorial
    pixi install --locked -e default
    pixi run --locked -e default test
    pixi run --locked -e default path-inference

    If you extracted the ZIP elsewhere, change the first line to the path of its tutorial folder.

Linux (x86-64)

Use a 64-bit Intel or AMD Linux system. ARM Linux is not included in the supplied lockfile.

  1. Install Pixi. Open a terminal and run:
    curl -fsSL https://pixi.sh/install.sh | sh

    Close and reopen your terminal, then check that pixi --version works.

  2. Download the tutorial. Download the repository ZIP and extract it into your Downloads folder. You should have a folder named sbi4finance-main.
  3. Install, check and open a notebook. In the new terminal, run:
    cd ~/Downloads/sbi4finance-main/tutorial
    pixi install --locked -e default
    pixi run --locked -e default test
    pixi run --locked -e default path-inference

    If you extracted the ZIP elsewhere, change the first line to the path of its tutorial folder.

Windows (x86-64)

The lockfile includes 64-bit Windows, but this tutorial has not yet been runtime-tested on Windows.

  1. Install Pixi. Open PowerShell and run:
    powershell -ExecutionPolicy Bypass -c "irm -useb https://pixi.sh/install.ps1 | iex"

    Close and reopen your terminal, then check that pixi --version works.

  2. Download the tutorial. Download the repository ZIP and extract it into your Downloads folder. You should have a folder named sbi4finance-main.
  3. Install, check and open a notebook. In the new terminal, run:
    cd "$env:USERPROFILE\Downloads\sbi4finance-main\tutorial"
    pixi install --locked -e default
    pixi run --locked -e default test
    pixi run --locked -e default path-inference

    If you extracted the ZIP elsewhere, change the first line to the path of its tutorial folder.

The marimo editor opens in your browser. Stop it with Ctrl + C before opening another notebook. The Heston and rough Bergomi notebook runs from the lab setup.

Already have the repository? Run the installation commands from its tutorial folder. See the tutorial setup guide and Pixi installation instructions for more detail.