Learning Systems Group

Computational Finance Research

Interest rate modelling

A debt instrument involves a number of payments of interest in the future. Therefore, to value the instrument, a realistic model for the dynamics of interest rates and an accurate estimation of the parameters of that model are needed. We use the concept learning from hints to obtain more accurate estimates of the parameters. We have also developed models to estimate the forward interest rate from observed market prices, which are very useful in the valuation of certain interest rate derivatives.

Volatility estimation

The volatility is a crucial parameter for financial markets. It enters as a parameter in pricing formulas for derivatives instruments, and plays a key role in asset allocation and risk management. As a result, it is important to be able to estimate accurately from the data. We have conducted a study that analyzes estimates based on the maximum likelihood approach. We have developed more accurate estimates that are based on using the high and the low prices in addition to the closing prices. These are readily available in market data series and lead to more accurate estimates.

American option valuation

American options generally do not have closed form solutions, and also their numerical evaluation is very tedious. We are currently developing numerical methods based on neural networks. The advantage of using neural networks is that once trained, they would be able to evaluate the option price very quickly.