Nonlinear Time Series Analysis via Neural Networks
This article deals with a time series analysis based on neural networks in order to make an effective forex market [Moore and Roche, J. Int. Econ. 58, 387–411 (2002)] pattern recognition. Our goal is to find and recognize important patterns which repeated
- PDF / 135,952 Bytes
- 4 Pages / 439.36 x 666.15 pts Page_size
- 64 Downloads / 255 Views
bstract This article deals with a time series analysis based on neural networks in order to make an effective forex market [Moore and Roche, J. Int. Econ. 58, 387– 411 (2002)] pattern recognition. Our goal is to find and recognize important patterns which repeatedly appear in the market history to adapt our trading system behaviour based on them.
1 Introduction Market systems are chaotic systems by their nature. One of the essential characteristics of a chaotic system is its extreme sensitivity to initial conditions. A tiny change in values at the beginning of the time series produces drastic changes in behavior later on. These systems are hard to predict. What we would like to achieve is to adapt our trading system behavior in accordance with important events in the market characterized by well known recognized patterns. Patterns in the market are not exact; they are slightly different every time they appear. They can have different amplitude and different duration, albeit visually the same pattern can look differently despite being the same. Patterns can be seen as some sort of maps that helps us to orientate in certain situations and navigate us to profitable trades. We choose several typical continuation and reversal patterns to describe with emphasis on the reliability of pattern recognition in the time series. For all chosen patterns we have to develop and optimize an input pattern form, i.e. how to read and record input patterns. Such optimized inputs also reduce calculation costs. Then we would be able to classify them and recognize similar patterns as well.
E. Voln´a () M. Janoˇsek V. Kocian M. Kotyrba Department of Informatics and Computers, University of Ostrava, Dvoˇra´ kova 7, Ostrava, 702 00, Czech Republic e-mail: [email protected]; [email protected]; [email protected]; [email protected] S.G. Stavrinides et al. (eds.), Chaos and Complex Systems, DOI 10.1007/978-3-642-33914-1 58, © Springer-Verlag Berlin Heidelberg 2013
415
416
E. Voln´a et al.
These input patterns do not cover every time point in the series, but are optimized to be suitable candidates in experimental tasks so that the developed classifiers would be able to learn key characteristics of these patterns and accurately recognize them.
2 Pattern Recognition Recent studies show that market patterns might implicate useful information for stock price forecasting. For the last few decades, neural networks have shown to be a good candidate for solving problems with the market analysis. We use Hebb’s neural networks with the delta learning rule [1]. Market patterns can be classified into two categories: continuation patterns and reversal patterns. Continuation patterns indicate that the market price is going to keep its current movement trend; while reversal patterns indicate that the market price will move to the opposite trend. The whole pattern recognition process consists of several steps which include data preparation, the pattern recognition itself and results utilization. The data preparation process makes the bigge
Data Loading...
