Comparative Analysis Of Machine Learning Algorithms For Cryptocurrency Price Prediction

Today’s economy needs transactions to be fast,cheap,and reliable.One of the innovative and seemingly strange tools for transactions and payments is the creation of cryptocurrency in 2008 by pseudonymous Satoshi Nakamoto.Bitcoin,the first cryptocurrency,is a decentralized network used to make private,anonymous,and peer-to-peer transactions anywhere worldwide.On the other hand,the price prediction problem is essential for both financial analysts and traders.Moreover,machine learning approaches recently emerged in time series prediction problems.Therefore,this study aims to move with the flow using new machine learning methods for the BTC price prediction problem.Furthermore,policymakers,central bankers,portfolio managers,investors,and traders need to identify the short-term and longterm BTC price predictive variables.Even though few studies in the existing literature have investigated the short-term or long-term predictability power of macroeconomic,microeconomic,blockchain information,and technical indicators,there is still hesitation between authors if macroeconomic and blockchain information attributes have long-term or short-term predictability power.Moreover,many of the existing works on BTC price predictions are empirical analyses without enough theoretical support.This study will investigate whether the macroeconomic,microeconomic,and blockchain information indicators based on economic theories predict the BTC price.Furthermore,there is not enough available literature on the BTC price prediction problem with machine learning compared to stocks.Still,there is an argument among authors regarding the superiority of machine learning techniques on conventional methods.Hence,more research should be conducted to show whether machine learning algorithms are superior to traditional methods.With taking these considerations into account,the following questions are raised:(1)What are the significant variables as short-term or long-term BTC price predictors?(2)What are the underlying economic theories of BTC price predictors?(3)Are machine learning algorithms superior to traditional methods for BTC price prediction? What machine learning model performs better? What are the best feature selection techniques?To answer research questions,this study applied several comparative analyses to the BTC price prediction model on different BTC data sets to verify the underlying theoretical analysis and answered the research questions.The empirical analysis consists of four parts: in the first part,a comparative analysis including ARIMA,GARCH,ANN,NARX,ANFIS,and SVR models is applied to the historical BTC price from January 1,2013,to January 1,2020,for one day,two days,and four days ahead prediction.The outcomes found that the historical BTC price is a significant short-term technical indicator confirming that technical analysis explains BTC price movements.In the second part,a multi-linear regression model is employed on monthly BTC prices from August 18,2010,to September 17,2018,considering some macroeconomics and blockchain information variables as independent variables.The results indicated that the mentioned variables are significant long-term predictors proving that supply & demand theory and cost-based pricing theory are underlying theories of BTC price predictors.In the third and fourth parts,two types of comparative analysis compare OLS,Ensemble methods,SVR,and MLP models on different datasets(BTC daily data set from October 11,2016,to June 12,2017,and BTC daily data set from January 1,2018,to June 5,2018).The findings showed that some technical indicators are significant short-term BTC price predictors confirming that technical analysis can predict the short-term BTC price as it was expected.Moreover,the overall findings showed that SVR is superior to other machine learning models and traditional models.Furthermore,no feature selection technique is proven to be the best.The findings of the current research are significantly impactful in academia and outside of academia.In academia,this study provides a foundation in BTC price prediction upon which further studies can be conducted.For outside academia,including policymakers,central bankers,traders,investors,and portfolio managers,this research provide some policy suggestions and recommendations,such as using technical indicators for shortterm and macroeconomic and blockchain information indicators for long-term BTC prediction problems.Moreover,it is recommended for machine learning developers and financial time series prediction experts to use emerging machine learning models like SVR for prediction purposes instead of traditional methods.