Applications of tensor neural networks to financial forecasting in incomplete markets

pre-agreed price. The problem of what an option is worth is usually solved using costly numerical simulation methods.
We will apply Deep Learning methods to solve the Option Pricing Problem. This cutting edge technique has the advantage that, once trained, the model can simulate many scenarios at a low computational cost. Unfortunately, training is costly and can be unstable.
Our original contribution is to use Tensor Networks to improve this Deep Learning model. Tensor Networks are known to give an advantage when training Neural Networks, and allow for more robust training, even when the market data is incomplete.
The partner will benefit from this project by developing their software stack to tackle this commercially valuable problem.

Faculty Supervisor:

Chi-Guhn Lee

Student:

Raj Gaurangbhai Patel

Partner:

Multiverse Computing

Discipline:

Engineering - mechanical

Sector:

Professional, scientific and technical services

University:

University of Toronto