Bitcoin Crypto-Asset Prediction: With an Application of Linear Autoregressive Integrated Moving Average Method, and Non-Linear Multi-Layered and Feedback Artificial Neural Network Models

The aim of the study is to evaluate two commonly used time series methods for forecasting Bitcoin (BTC) prices: the Autoregressive Integrated Moving Average (ARIMA) and the Multilayer Perceptron (MLP) Neural Network. The dataset consists of weekly BTC values from the period between 2020 and mid-2022, containing a total of 135 observations. The analyses were conducted using R-Studio software. The stationarity of the data was checked using ADF, PP, and KPSS unit root tests. For predictions, a linear method, ARIMA, and a nonlinear method, the MLP Neural Network model, were utilized. Although the MLP model demonstrated better performance, it is challenging to indicate a definitive superiority due to its limitations. The relatively low performance of both models may be attributed to the extreme volatility and speculative nature of cryptocurrencies, along with their tendency to behave independently of their underlying structures.