High-Frequency Trading of Cryptocurrencies Through Short-Term Cointegration Pairs-Trading Strategies

This paper focuses on estimating long-term relations of cryptocurrency exchange rates and trading based on short-term deviations from this long-run equilibrium. These shortterm deviations are exploited by means of statistical arbitrage strategies incorporating cointegration methods. Cointegrating relations in the cryptocurrency market are analyzed, sorted and chosen by means of the Engle-Granger two-step method (EG2SLS). Additionally, we perform a Johansen test for cointegration on the resulting EG2SLS cointegrating relations. Next, we form a pairs-trading portfolio based on the estimated cointegrating relations. A highfrequency trading bot is then configured to trade with this portfolio. We then compare the cointegration-based pairs-trading performance with and without the additional Johansen test step. The weekly performance of both cointegration-based pairs-trading methods is historically simulated on 16 weeks of cryptocurrency data. We conclude that the Johansenassisted market neutral pairs-trading strategy outperforms its EG2SLS counterpart. The Johansen and EG2SLS pairs-selection methods result in a weekly return of 6.81% and 5.97%, respectively, including representative transaction costs. To decrease future computation time, we also consider a method to find a set of exchange rates with a higher probability of being cointegrated. We estimate lead/lag relations of exchange rates in the cryptocurrency market by means of Granger causality tests. We conclude that the lagging exchange rates are more likely to be cointegrated. This means that we can define a subset of all exchange rates, which are more probable of being cointegrated.