Neural network and reduced order modeling for financial risk analysis

Abstract

Financial risk analysis is an excellent way of identifying and assessing the factors that may negatively affect investments. Investors use it to determine whether to undertake a particular investment, the plausible return and how to mitigate an activity’s potential losses. However, the risk analysis of financial instruments often requires the valuation of such instruments under a wide range of future market scenarios, demanding efficient algorithms. Thus, we propose an alternative approach based on machine learning techniques for faster simulations with reliable outcomes. Instead of running costly numerical simulations, we construct a neural network model for the relation between the market scenarios and the instrument values. We generate the training data by solving the financial instrument using an appropriate valuation function for some market scenarios. The trained neural network can then be used to solve the financial instrument for the remaining scenarios. Since we perform only a limited number of full valuations and the rest of the valuations are based on the trained neural network, we can achieve our necessary speedup.