Research On The Forecasting Method Of The Reversal Point Of Digital Currency Price

In recent years,the digital currency represented by BTC and the related blockchain technology have been spread all over the world,and the upsurge of "blockchain" and "digital currency" has been aroused.The United States and some European countries have legalized the digital currency such as BTC,and more and more domestic and foreign investors have begun to enter this new field.Digital currency can be used for both investment speculation and payment transactions.Therefore,the analysis and prediction of the trend of digital currency price has become the focus of the investment community.But digital currency is an emerging market.Compared with the stock market,it is very different.All kinds of factors will cause great market volatility.Compared with the specific price forecast,the price fluctuation based on the trend is more realistic and easier for investors to grasp.Therefore,based on the analysis of technical indicators and the related machine learning algorithm,this paper constructs a data model for the prediction of the reversal point of digital currency.(1)This paper studies the influence of different data processing methods on the statistical accuracy of a single technical index.It is the most important way for investors to judge the reversal point of digital currency directly with the technical index value,which is also a simple and effective method.However,the existing results of digital currency price forecasting have not paid attention to the statistical accuracy of single technical indicators.In the existing research results on digital currency price,the processing of the original data is basically standardized,logarithmic processing,etc.This paper uses the signal extraction processing model of other disciplines for reference to process the original data of digital currency price.This paper selects 7 representative digital currencies in the top 20 digital currencies with market value,intercepts 1638 data from August 7,2015 to January 30,2020 as samples,and processes the original data in three ways:(1)no processing;(2)standardized processing;(3)FastICA processing.Then calculate the technical indicators of three treatment methods,such as MACD,RSI,bias,KDJ and real reversal point vector construct the data set of price reversal point based on time series;compare several technical indexes of three cases with real reversal point respectively,analyze the accuracy rate that each technical index can achieve;find that bias is the single technical index with the best comprehensive performance,and the statistical accuracy rate of single technical index after FastICA processing is the highest.(2)On the basis of establishing the data set of technical index reversal point,this paper uses SVM three classification model to predict the price reversal point.The experimental results show that the statistical accuracy of a single technical index is generally low,which can not achieve a better investment guidance.The accuracy of SVM model is much higher than that of single technical index.Moreover,the prediction accuracy of FastICA-SVM model is higher than that of other two data processing methods.The necessity of using SVM model and the superiority of FastICA data processing method over the existing data processing method are explained.(3)Based on the data processing of FastICA,the imbalance data is processed and the prediction accuracy is analyzed.Firstly,the data sets of up and down reversal points are constructed,and the problem of data imbalance is found in both data sets.Then handle the two unbalanced data sets with SMOTE model,establish a two classification prediction model based on SVM and evaluate the prediction accuracy of the model.It is found that the prediction accuracy of FastICA-SMOTE-SVM model is higher than that of the inversion point of FastICA-SVM model.(4)Compare the cumulative yield of FastICA-SMOTE-SVM model and FastICA-SVM model after backtesting under specified trading rules.It is found that the FastICA-SMOTE-SVM model has better overall effect,and the two models are in the "bear market" stage.The innovation of this paper is as follows:(1)the research object is changed from the specific price forecast to the reversal point forecast,which reduces the difficulty of work and has more guiding significance for investors.(2)The improved method based on independent component analysis(FastICA),which is used in engineering for mixed signal separation,is applied to the decomposition and research of digital currency price time series,and the trend prediction of digital currency price is studied from a new perspective.(3)Considering the data imbalance of digital currency price reversal point,the SMOTE tool is used to optimize the sampling,avoiding the migration of the model to the large class and affecting the final effect.(4)In this paper,we not only evaluate the model,but also carry out simulation test through backtesting experiment,and compare different models based on the cumulative yield over time,which makes the model comparison more convenient.