Forecast Of Foreign Exchange Rate Trend Based On Time Series Similarity

Quantitative trading has a history of more than 40 years in developed countries of Europe and America.With its good performance,it has been recognized by more and more investors.Quantitative trading makes investment strategies based on historical data,mathematical model,powerful computing power of computer.We can find that there are always many similarities among historical data.Therefore searching similar trends in history has great significance for trading.In view of the existing time series research technologies at home and abroad,This paper put up a modol to predict future data based on KNN.Then the related parameters were studied by experiment.The results show that this modol works well on foreign exchange price especially improvement modol.The main contents and contributions are as follows:1.In view of the existing time series research technologies at home and abroad,we highlight two key technologies—dimension reduction and similarity measurement,and summarize common dimension reduction and similarity measurement methods.Considering the characteristics of foreign exchange price data,such as large fluctuation and obvious trend characteristics,we study the dimension reduction algorithm based on trend points experimentally,preparing the research foundation for the follow-up work.Results show that the algorithm can retain the trend of foreign exchange prices well and is suitable as a preprocessing method for foreign exchange price data.2.We design a foreign exchange price prediction method based on KNN algorithm.Firstly,the historical foreign exchange price data are preprocessed,such as normalization,dimension reduction,linear interpolation,etc.Then,the KNN algorithm is used to screen out the most similar historical data and vote to predict the future price trend.The real foreign exchange price data in a certain time period are selected for experiments,and the influences of similarity calculation method,number of voting rights and dimension reduction operation on prediction accuracy are evaluated respectively.The experiment show that the above processing methods can effectively screen out similar foreign exchange prices.On this basis,statistical prediction based on KNN principle can achieve high accuracy.3.We propose a foreign exchange price trend prediction method based on convolutional neural network.On the basis of similar sequences,three kinds of technical indicators,namely moving average,overbought and oversold,trend and volume,are selected to transform one-dimensional similar sequence data into two-dimensional data and train the convolutional neural network.We select the real foreign exchange price data in a certain time period for experiments,and evaluate the influences of the size and number of convolution kernels,and the number of full connection layers on the prediction accuracy respectively.The comparsion between the prediction method based on similar sequences and prediction method based on convolutional neural network is emphasized.Results show that the use of convolutional neural network based on similar sequences effectively improves the prediction accuracy of foreign exchange price trend.