Hybrid Neural Systems in Exchange Rate Prediction
In this chapter, a new hierarchical hybrid wavelet — artificial neural network strategy for exchange rate prediction is introduced. The wavelet analysis (the Mallat’s pyramid algorithm) is utilised for separating signal components of various frequencies a
- PDF / 332,458 Bytes
- 20 Pages / 439.2 x 666 pts Page_size
- 35 Downloads / 195 Views
Institute of Computer Science, Jagiellonian University, Krak´ow, Poland. [email protected] Motorola Polska, Krak´ow, Poland. [email protected] AGH University of Science and Technology, Krak´ow, Poland. [email protected]
Summary. In this chapter, a new hierarchical hybrid wavelet - artificial neural network strategy for exchange rate prediction is introduced. The wavelet analysis (the Mallat’s pyramid algorithm) is utilised for separating signal components of various frequencies and then separate neural perceptrons perform prediction for each separate signal component. The strategy was tested for predicting the US dollar/Polish zloty average exchange rate. The achieved accuracy of prediction of value alterations direction is equal to 90%.
12.1 Introduction Generally, two hypothesis concerning predictability of markets exist. The first one states that it is impossible to predict market behaviour whereas according to the second hypothesis, markets can be described by their own statistical dynamics modelled by walk-type processes with a memory (see (28)). Generally, a market which is weak dependent on political decisions, ecological catastrophes, which has a great number of participants, a great inertia of its processes, is usually, partially predictable. There are several tools, based on statistical methods, dynamical systems theory, approximation theory and artificial intelligence, used for forecasting time-series behaviour. The review of these methods, areas of applications, and obtained results, can be found in (40). In this chapter, time series prediction of the currency market is considered. The specifics of the currency market is discussed in detail in section 12.3. The aim of this chapter is to present results obtaining by using a hybrid artificial intelligence (AI) system for forecasting of the Polish zloty - US dollar exchange rate. A hybrid artificial neural network (ANN) - wavelet analysis system is used. In the next section the specifics of hybrid AI neural systems is discussed. Section 12.3 is devoted to describing the currency market and previous literature on currency market prediction. Later sections describe the methodology used and the results obtained - see also (10). Finally, some conclusions are presented.
A. Bielecki et al.: Hybrid Neural Systems in Exchange Rate Prediction, Studies in Computational Intelligence (SCI) 100, 211–230 (2008) c Springer-Verlag Berlin Heidelberg 2008 www.springerlink.com
212
A. Bielecki et al.
12.2 Neural Networks and Hybrid Systems An artificial neural network is a cybernetic system whose structure and activity is modelled after animals’ and humans’ nervous systems, in particular brains. A neuron is a basic signal processing unit. The first neural model, very simplified in comparison with a biological neural cell, was described in 1943 (see (25)). According to this model, a neuron is a module having a few weighted inputs and one output see fig.12.1. Input signals, say x1 , ..., xM , and weights w1 , ..., wM constitute vectors x = [x1 , ..., xM ] ∈ RM and w
Data Loading...
