Predicting changes in the federal funds target rate using artificial neural network

Abstract

The determination of federal funds target rate is arguably the most important decision for the Federal Reserve. As it is a complex decision making process, with the exact functions unknown, it is always difficult for a researcher, using traditional statistical methods, to model the thoughts of the FOMC, and hence, predict the change in the federal funds target rate. However, with neural networks fast evolving as a promising methodology, we may finally be able to emulate the FOMC's decision making. In fact, neural networks have already been successful in various business fields, ranging from marketing (Dasgupta et. al., 1994) and tourism demand analysis (Uysal and Roubi, 1999), to stock picking (Donaldson and Kamstra, I996) and derivative applications (Grudnitski and Osburn, 1993). The aim of this paper is to apply established neural network architectures on the forecasting of change in the federal funds target rate. The period of study is during the term of Chairman Greenspan, where the Fed emphasizes on a large set of economic time series to make their decisions. Prediction using different window sizes and for different horizons will be studied. 2 validation schemes will be used to separately account for possible change in the underlying function and sample size. A successful application of the neural networks will have widespread benefits to global financial market players.