Currency Predictions for Multi-Currency Instruments

Multi-currency instruments are quite popular in Iceland. This stems from the fact that loans based on the Icelandic Krona have high interest rates. Therefore, corporations and corporate customers often prefer to take loans in foreign currencies. In order to minimize the currency risk it is common to construct a loan that consists of baskets of currencies. In this study, we implement various neural network training algorithms as well as construct a mathematical model in an attempt to predict future currency rates. Implemented models are compared based on accuracy and performance with respect to forecast periods, input models and training data partitions. These predictions can then be used to compose an optimal set of baskets for a given loan.