| Summary: | 碩士 === 元智大學 === 資訊管理學系 === 107 === Since the birth of Bitcoin in 2009, the market for virtual currencies has been built and expanded continuously. Accompanying with the rise of this market, it also attracts people to invest the activity to “mine” the virtual currencies. As a result, a lot of real currencies have been threw into this market and push the price of Bitcoin, the earliest mined and the most known virtual currency, to reach the historical high on December 16, 2017. At that time, the price per Bitcoin was list as $19,665.39. However, the price of this virtual currency began to fall in January 2018. Recently, its price is swinging around $3,000 to $4,000, about 15-20% from its peak. Now that the price of virtual currencies, such as the Bitcoin, fluctuate frequently, their predictions are important because those can help the investors to make the correct investing decision.
This research has two objects. Firstly, the methods of machine learning are used to predict the ups and downs of the cryptocurrency. Second, the time series model is used to predict the price trend of the cryptocurrency. The methods of machine learning applied here are the RandomForest, SVM, KNN, and LogisticRegression in Python language to predict the most suitable algorithm for each currency. We also apply the autoregressive integrated moving average model (ARIMA) to predict the trend of price for Bitcoin and Ethercoin. The data we use in this study is collected from CoinGeco within the period of 2013 to 2018. By executing the methods of machine learning in Python and the EACF package in R, the best model to predict the ups and downs of cryptocurrency is selected and the best fitted ARIMA models were derived. Then we apply this best fitted model to forecast the price trend of cryptocurrency, including the intra-sample and out-of-sample forecasts.
|
|---|
