Studying the patterns and long-run dynamics in cryptocurrency prices

This study analyses the price movements of a select sample of cryptocurrencies and examines whether they are cointegrated and predictable using machine learning algorithm and Johansen Test. The study used daily historical trading data of 76 cryptocurrencies sourced from different cryptocurrency exchanges. A sub-sample of six cryptocurrencies were chosen for the cointegration and machine learning analysis based on their market share, attractiveness to the investors and availability of data for the full sample period. The data records starting from April 29, 2013 to February 7, 2019 were considered for the study. An error correction model was estimated to investigate both the long-run and short-run dynamics between the cryptocurrency prices. The evidence from the error correction model estimates shows that there is a long-run association between the prices of crypto currencies. The machine learning algorithm involving neural networks (multilayer perception) was used to comprehend the data patterns in the cryptocurrency price series, and the results show that the model fits well in identifying and predicting the data patterns. The study also examines the possible value drivers of cryptocurrencies by estimating a linear regression with a set of covariates, which include the cryptocurrency demand and supply interaction variables and financial variables such as the NZX/S&P 50 index and exchange rates. The linear model estimates confirm that cryptocurrency market fundamentals have an important impact on cryptocurrency prices; however, they do not support the prediction that financial fundamentals are the major value drivers of cryptocurrencies.