Hidden Markov Models for Cryptocurrency Trading

Hidden Markov models is a statistical signal prediction model for sequential data, which has been widely used, among else, to predict economic regimes and stock prices. In this work, we introduce the application of various extensions of classical Hidden Markov Models in cryptocurrency trading based on the market state predictions. We first introduce the basic concepts of Hidden Markov Models and the Baum-Welch algorithm for parameter estimation along with the Viterbi algorithm for state prediction. We consider several models with different assumptions on the distribution of the observations, including the Categorical, Gaussian and Gaussian-Mixture Model. Furthermore, we present the theory and application behind Hidden semi-Markov Models and Parametric Hidden Markov Models. Having estimated the model parameters on historical data, we then apply the models to predict the market regimes of the cryptocurrency market and use these predictions alongside predefined trading strategies to trade on the Binance spot market. Lastly, we evaluate the performance of the models based on their average total return on multiple sequences and compare it return with the benchmark strategy of buy-and-hold.