| Summary: | Bitcoin has recently received special attention in economics and finance as the most popular blockchain technology. This dissertation aims to discuss whether newly machine-leaning models perform better than traditional models in forecasting. Particularly, this study compares the accuracy of the prediction of bitcoin prices using two different models: Long-Short Term Memory (LSTM) versus Auto Regressive Integrated Moving Average (ARIMA), in terms of forecasting errors, and Python routines were used for such purpose. Bitcoin price time series ranges from 2017-06-18 to 2019-08-07, in a daily basis, sourced from the Federal Reserve Economic Data. To compare the results of both models, data was divided into two subsets: training (83.5%) and testing (16.5%). The literature usually indicates that LSTM outperforms ARIMA. In this dissertation, the results do confirm that LSTM forecasts of bitcoin prices improve on average ARIMA predictions by 92% and 94%, according to RMSE and MAE.
|
|---|