LIBOR Prediction Using Genetic Algorithm and Genetic Algorithm Integrated with Recurrent Neural Network

Abstract

LIBOR (London Inter Banking Offered Rate) market, as one of the most important interest rate market to display the liquidity of the currency flow globally, is a huge market with global stake holders to be involved. The rate is a significant indicator for the liquidity risk of global markets in different currencies and for the stability of the finance markets in main countries. This thesis is to predict the LIBOR in USD with two algorithms: the pure Genetic Algorithm ('GA') and the Genetic Algorithm integrated with Recurrent Neural Network ('GA + RNN'). The pro and cons of these two algorithms will be compared, with fitness values and Mean Squared Error ('MSE'). 50 test cases will be created and tested randomly to verify the performance of the predictions. In the algorithm of RNN+GA, the target variance between predication and real value is no more than 0.015.