Comparing market phase features for cryptocurrency and benchmark stock index using HMM and HSMM filtering

Abstract

A desirable aspect of financial time series analysis is that of successfully detecting (in real time) market phases. In this paper we implement HMMs and HSMMs with normal state-dependent distributions to Bitcoin/USD price dynamics, and also compare this with S&P 500 price dynamics, the latter being a benchmark in traditional stock market behaviour which most literature resorts to. Furthermore, we test our models’ adequacy at detecting bullish and bearish regimes by devising mock investment strategies on our models and assessing how profitable they are with unseen data in comparison to a buy-and-hold approach. We ultimately show that while our modelling approach yields positive results in both Bitcoin/USD and S&P 500, and both are best modelled by four-state HSMMs, Bitcoin/USD so far shows different regime volatility and persistence patterns to the one we are used to seeing in traditional stock markets.