Papers
Topics
Authors
Recent
Search
2000 character limit reached

Applying Deep Learning to Calibrate Stochastic Volatility Models

Published 14 Sep 2023 in q-fin.CP, cs.AI, q-fin.MF, q-fin.PR, and q-fin.RM | (2309.07843v2)

Abstract: Stochastic volatility models, where the volatility is a stochastic process, can capture most of the essential stylized facts of implied volatility surfaces and give more realistic dynamics of the volatility smile/skew. However, they come with the significant issue that they take too long to calibrate. Alternative calibration methods based on Deep Learning (DL) techniques have been recently used to build fast and accurate solutions to the calibration problem. Huge and Savine developed a Differential Machine Learning (DML) approach, where Machine Learning models are trained on samples of not only features and labels but also differentials of labels to features. The present work aims to apply the DML technique to price vanilla European options (i.e. the calibration instruments), more specifically, puts when the underlying asset follows a Heston model and then calibrate the model on the trained network. DML allows for fast training and accurate pricing. The trained neural network dramatically reduces Heston calibration's computation time. In this work, we also introduce different regularisation techniques, and we apply them notably in the case of the DML. We compare their performance in reducing overfitting and improving the generalisation error. The DML performance is also compared to the classical DL (without differentiation) one in the case of Feed-Forward Neural Networks. We show that the DML outperforms the DL. The complete code for our experiments is provided in the GitHub repository: https://github.com/asridi/DML-Calibration-Heston-Model

Summary

  • The paper demonstrates a Differential Machine Learning approach that integrates both price and gradient data to efficiently calibrate the Heston model.
  • It leverages a twin network architecture with differentiable activations to enhance calibration accuracy and reduce computation time from hours to seconds.
  • Experiments on synthetic and S&P 500 options data validate its real-time applicability and superior performance compared to traditional methods.

Applying Deep Learning to Calibrate Stochastic Volatility Models

Overview

The paper addresses the challenge of efficiently calibrating stochastic volatility models for financial markets, particularly focusing on the Heston model. These models provide a more realistic depiction of market dynamics compared to deterministic models but suffer from extensive computational demands. The authors propose a Differential Machine Learning (DML) approach to improve calibration speed and accuracy, utilizing deep neural networks (DNNs) to approximate pricing functions and their sensitivities.

Differential Machine Learning Approach

The DML approach involves the training of deep neural networks not only on option price data but also on the differentials of these prices with respect to input parameters, enhancing the network's ability to generalize and provide accurate pricing predictions with fewer data. This technique leverages backpropagation to calculate gradients efficiently and integrate them into the learning process.

  • Architecture: The paper employs a twin network architecture that predicts both option prices and their gradients, sharing weights between the forward and backward propagation phases. Activations in this network are chosen to be differentiable, ruling out functions like ReLU.
  • Training: The cost function combines the mean squared error (MSE) of prices and their differentials, weighted to prevent overfitting noisy labels. This serves as a regularization technique, akin to data augmentation, by incorporating more training data without bias introduction.

Numerical Experiments and Results

The experiments demonstrate the effectiveness of the DML technique using synthetic data generated by a range of model parameters. Regularization techniques such as dropout and gradient clipping are explored, with DML showing superior performance in terms of calibration accuracy and computational efficiency.

  • Comparison with Traditional Methods: Differential evolution, a method traditionally used for global optimization, is compared with the DML approach. While both methods achieve similar calibration accuracy, the DML method drastically reduces computation time from hours to seconds.
  • Application on Real Data: The trained DNN is tested against S&P 500 index options data, showing that DML-calibrated models match market prices closely and are computationally efficient, making real-time calibration feasible.

Implementation Details

For reproducibility, the paper provides pseudocode and implementation strategies:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
import torch
import torch.nn as nn
import torch.optim as optim

class DifferentialNN(nn.Module):
    def __init__(self):
        super(DifferentialNN, self).__init__()
        self.fc1 = nn.Linear(input_size, 50)
        self.fc2 = nn.Linear(50, 50)
        self.fc3 = nn.Linear(50, output_size)
        self.activation = nn.Softplus()

    def forward(self, x):
        z1 = self.activation(self.fc1(x))
        z2 = self.activation(self.fc2(z1))
        y = self.fc3(z2)
        return y

model = DifferentialNN()
optimizer = optim.Adam(model.parameters(), lr=0.001)
criterion = nn.MSELoss()

for epoch in range(num_epochs):
    for inputs, labels in data_loader:
        optimizer.zero_grad()
        outputs = model(inputs)
        loss = criterion(outputs, labels)
        loss.backward()
        optimizer.step()

Conclusion

The DML framework offers a scalable and efficient solution for calibrating complex financial models like the Heston model, providing both theoretical rigor and practical applicability. Future work may focus on extending DML to other financial instruments and models, potentially integrating more sophisticated deep learning architectures to further enhance performance and utility in a production environment.

Paper to Video (Beta)

No one has generated a video about this paper yet.

Whiteboard

No one has generated a whiteboard explanation for this paper yet.

Open Problems

We haven't generated a list of open problems mentioned in this paper yet.

Authors (2)

Collections

Sign up for free to add this paper to one or more collections.