Predicting Short-term Foreign Exchange Based On Wavelet Neural Network

As an international capital speculation places, the history of foreign exchange market is much shorter than stocks, gold, interests rate and futures market. However, it is rapidly growing at a surprising speed. According to statistics, the daily transaction volume in the foreign exchange is more than 2 trillion U.S. dollars, which far exceeds other financial products markets such as stocks and futures. It has become the largest market in the world. The largest foreign exchange markets are in London, New York and Tokyo respectively. The foreign exchange market is the inevitable product of economic development which can not be ignored on the part of economics in each country.The fluctuation of exchange rate always run with a trend:either upward, downward or consolidation that manifests the power balance between the demand and the supply. Take EUR/USD for an example, If it rises upward, this phenomenon shows the increasing demand for the EUR or the preference EUR to USD from the mainstream funds on the market. The reasons for this inclination are complicated. We cannot build a proper model to approach this nonlinear system accordingly. Nevertheless, the direction of future trend may be predicted by researching regulations of its movement, as once a trend runs, it can illustrate that one is stronger than the other, and it will normally continue for some time not to be reversed easily. Because the trend can maintain the original direction of the inertia,technical analysts will spend a lot of effort on finding out the law of price changes. In general,if the exchange rate has been a continuous increase or down in price over a period of time,there is no unexpected, exchange rate will continue to rise or fall in accordance with the direction of this price for some time to come, who has no reason to change the direction established.Therefore, we can study the history of the exchange rate price to discover the laws of exchange rate fluctuation and achieve the purpose of forecasting.This paper selects EUR/USD as a target market, combining with wavelet analysis and artificial neural network and then building a prediction model. Wavelet decomposition algorithm decomposes of the original price time series to a different frequency channel and reconstructs to the original scale, thus there has a single sequence that the frequency components is more single than the original. Then it makes use of neural network to predict the decomposed time series, and has a predictive value of the original time series by synthesizing finally.