Machine learning tools for rapid pricing of exotic equity products
Exotic derivatives of various kinds contribute significantly to the risk exposure that must be managed by banks. In order to be competitive, banks need to assess their risk exposure frequently, and make necessary adjustments to their positions. Assessing risk exposure involves computing valuations of all the assets in their investment portfolio, along with their sensitivities. When closed-form formulae are available, this can be done rapidly. However, no such formulae are available for exotic products, therefore various computationally demanding numerical methods must be used.
The goal of this project is to develop a machine learning (neural network) model that can replicate the valuations produced by BMO’s existing computational methodologies, given any feasible set of input parameters. Once trained, such a network will be able to generate its output rapidly. The challenge is in the training, since the space of input parameters is extremely high dimensional.
Antony Ware;Alexandru Badescu
Bank of Montreal
Mathematics
Finance and Insurance
University of Calgary
Accelerate