The academic paper realased by Blanka Horvath; Aitor Muguruza and Mehdi Tomas suggests to use deep learning as a speed up for pricing.
By learning the model from model parameters to expected payoff, we can relocate the time-consuming part of calibration, which is the functional evaluation of prices, to an offline pre-processing of the data. For the on-line calibration part, we are only calibrating a deterministic function, which is fast.
Neural networks efficiently approximate complex pricing functions offline, addressing calibration challenges in models with slow pricing. Such as Rough Volatility models like the rBergomi model.
In this repo we implement the paper with a c++ library to speed up the online pricing and we put the off-line learning phase in a time-laps in order to visualise it. We present a graph with the calibration duration.
repo.mov
The main file:
- File: main.ipynb
- Execute the entire program using this Python script
The C++ File:
- File: cppNeuralNet.cpp
- Includes the hardcoded Neural Net and its gradient
Compilation Output:
- Output File: cppNeuralNet.cp311-win_amd64.pyd
- Facilitates easy code utilization without the need for C++ code compilation.
Configuration File:
- File: setup.py
- Used for C++ compilation if required.
Data File:
- File: TrainrBergomiTermStructure.txt.gz
- Used to train the Neural Network
Time-laps.mov
Python 3.11.6
Eigen 3.4.O
g++.exe (Rev2, Built by MSYS2 project) 13.2.0
Compilation if needed : python setup.py build_ext --inplace don't forget to relocate :'C:/Users/AD/Desktop/testcpp/eigen/eigen-3.4.0' inside the setup.py
MIT License
A short and simple permissive license with conditions only requiring preservation of copyright and license notices. Licensed works,
modifications, and larger works may be distributed under different terms and without source code.