The Study Of Exchange Rate Prediction Based On BP Neural Network

The change of exchange rate is unpredictable, it can not only affect the individual’s life, the growth of the company, but also affect a country’s economy. People have been looking for a way to the fair and stable exchange rate for one thousand years, today people are still exploring.Exchange rate changes frequently, because it will be affected by many factors: the international balance of payments, the surplus and deficit, inflation, monetary purchasing power, interest rate level, national macro-control, economic policy, foreign exchange, the influence of government intervention and so on, they are affecting the exchange rate, and promoting the changes of exchange rate.How to predict exchange rate quickly and effectively, economists have explored this problem for a long time, because exchange rate is affected by many factors, showing irregular nonlinear variation trend, so it bring great difficulties to its prediction. The neural network is a good way to predict nonlinear data, so it can provide a realistic and effective method for the prediction of exchange rate.First of all, this paper gives the stationary test about the selected data of RMB against the dollar, it shows that the exchange rate data is not stable, so I dispose the exchange rate data and then they are reposeful, after that I give the exchange rate volatility series’normality and BDS test, it verify the nonlinearity of exchange rate volatility series.BP neural network is widely used, this paper chose the method to predict the exchange rate of RMB and dollar. Although BP algorithm has many advantages and is widely used, it also has many shortcomings, such as that it is easy to fall into local optimum and convergence speed is slow, it is difficult to determine the number of input neurons, hidden layers and each layer neuron node.Finally in order to solve the above problems, this paper puts forward a improved BP algorithm. First, in order to solve the problem of BP algorithm is easy to fall into local optimum, I put forward the additional momentum method, in order to solve the problem of the neuron input number is difficult to determine, I chose the AC rule to determine the number of input neurons, and use MCPT to select the best sample set. The improved BP neural network algorithm is used to predict the exchange rate between RMB and dollar. And the forecast has carried on the inspection by using the error check index, the inspection results show that the improved algorithm is feasible.