Abstract
This research applies deep neural networks for pricing European options, evaluating
their performance under various pricing models: Variance Gamma, Heston,
Bates, and Variance Gamma with Stochastic Arrival model. The neural network
is trained using synthetic data computed using the COS and Brent’s iterative
method to calculate the price and implied volatility, respectively. Consequently,
the neural network learns the relationship between model parameters and volatility.
The results show the neural network can accurately price options with a small
margin of error compared to the COS method. Furthermore, the results show that,
although neural networks can accurately price, the computational time is fast but
still slower than the COS method, on average.