Forecasting the Prices of Cryptocurrencies Using GM(1,1) Rolling Model

Although cryptocurrencies initially emerged as a transnational payment instrument, it has become an investment tool by attracting the attention of investors within the functioning of the capitalist system. In this chapter, the use of cryptocurrencies as an investment tool rather than in commercial transactions is discussed. As is known, most investors remain in a dilemma between the risk of risk aversion and the maximization of returns. Investors in this dilemma try to predict the future price or returns of the financial instruments through various analyzes and thus make an effort to give direction to their investments. These analyses are generally carried out by analyzing the past values of prices or returns by adopting a technical analysis approach. However, since the cryptocurrencies are a relatively new investment tool, it is not possible to reach the previous period price and yield information for an extended period. For this reason, the scope of the chapter is to explain the functioning of the cryptocurrencies as an investment tool in the market and to share information about the types of investors who have transferred their funds to cryptocurrencies by providing statistical information. Then, it is aimed to share the theoretical knowledge about GM(1,1) Rolling Model which has been proved by the literature in which it produces successful results especially in forecasting problems in uncertainty environment. Finally, the price forecasting of popular cryptocurrencies which are Bitcoin, Ethereum, Litecoin, and Ripple was made using the GM(1,1) Rolling Model, and it was tested whether this model is advisable for price forecasting of cryptocurrencies. Results of the Model show that the forecasting errors ranged from 1.35% to 7.76% for 10-days period. Also, direction forecasting results are between 40% and 50% in the same period. Also, returns of the bitcoin investment which made by trusting the results are ranged from −0.60% to −8.18. The results may be considered that the model was successful in forecasting the prices but unsuccessful in the direction forecasting. Even though the estimates are made with low percentages, the time series analyzes made with the lagged data of Bitcoin prices are not successful. Therefore, the technical analysis approach can be interpreted as not sufficient for modeling Bitcoin prices. So, these results show that defining bitcoin price movements is not only a forecasting problem but also a classification problem.