Altcoins as alternatives for what?

After Bitcoin said hello world in 2009, several alternative digital currencies based on cryptographic technology (cryptocurrencies) were subsequently created. Those cryptocurrencies are called altcoins. This paper uses time-series data for altcoins to assess correlations among altcoins and Bitcoin (BTC), Euro (EUR), Japanese Yen (JPY), and Chinese Yuan (CNY). The dataset is a dynamic panel data (unbalanced) gathered from online content (https://coinmarketcap.com/currencies/views/all/) that provides real-time market data for more than 600 cryptocurrencies, but the data is not stored. To store the data, my research assistant accessed the web site almost daily between October 29, 2015 and December 24, 2015. At a certain time the collected data consists of 13,407 total observations with 540 altcoins and Bitcoin. For lag-2 regressions, effectively 359 altcoins with 7,402 observations are available.

For the analysis, we employ a nested dynamic model (a la simultaneous equations model) that controls endogeneity. The model is estimated by 2SLS. We then find that altcoins of larger market capitalizations (major altcoins) are positively correlated with EUR and negatively with CNY. Equivalent results are also obtained in the pooled regression model that uses a weight for BTC price based on relative market capitalization to treat altcoins of smaller market capitalizations (minor altcoins) as major ones. The weight is applied in order to adjust prices of altcoins to BTC price since they vary from few dollars to less than pennies. The regression for Bitcoin price also shows similar results: siginificantly positive correlation with EUR and negative with CNY. On the other hand, we also find that minor coins are negatively correlated with BTC. The results suggest that BTC is a substitute for minor altcoins and EUR and CNY are a complement and a substitute for major ones.

In the analysis, major and minor altcoins are classified by a structural breaking test. Identification tests, unit-root tests, and cointegration tests are provided for all regressions. In addition, heteroskedasticity is also considered and controled.