Exploring the Predictability of Cryptocurrencies via Bayesian Hidden Markov Models

Constandina Koki, Stefanos Leonardos, Georgios Piliouras

Abstract

In this paper, we consider a variety of multi-state Hidden Markov models for predicting and explaining the Bitcoin, Ether and Ripple returns in the presence of state (regime) dynamics. In addition, we examine the effects of several financial, economic and cryptocurrency specific predictors on the cryptocurrency return series. Our results indicate that the Non-Homogeneous Hidden Markov (NHHM) model with four states has the best one-step-ahead forecasting performance among all competing models for all three series. The dominance of the predictive densities over the single regime random walk model relies on the fact that the states capture alternating periods with distinct return characteristics. In particular, the four state NHHM model distinguishes bull, bear and calm regimes for the Bitcoin series, and periods with different profit and risk magnitudes for the Ether and Ripple series. Also, conditionally on the hidden states, it identifies predictors with different linear and non-linear effects on the cryptocurrency returns. These empirical findings provide important insight for portfolio management and policy implementation.

Introduction

Cryptocurrencies are notoriously volatile, but that volatility is not uniform over time. Markets often cycle through distinct phases—explosive growth, "crypto winters," and periods of relative stability. Traditional econometric models often assume a single, constant relationship between variables, which fails to capture these structural breaks. This study addresses the challenge of predictability by identifying hidden market states and determining if external factors—such as the S&P 500, gold prices, or cryptocurrency-specific metrics—drive the transitions between these regimes.

Methodology

The research evaluates several classes of models, primarily focusing on Hidden Markov Models (HMMs). These models assume the market is always in one of $m$ "hidden" states, each with its own mean return and volatility. The study highlights the Non-Homogeneous Hidden Markov (NHHM) model, which allows state switching probabilities to change based on external predictors, unlike standard homogeneous models.

Diagram of the Non-Homogeneous Hidden Markov Model (NHHM) architecture where predictors influence transition probabilities and hidden states determine returns.

Estimation was performed within a Bayesian framework using Markov Chain Monte Carlo (MCMC) methods. Key techniques included Polya-Gamma data augmentation for handling complex state transitions and a Reversible Jump algorithm for stochastic variable selection to identify the most relevant predictors.

Data and Variables

The study analyzed daily log-returns for Bitcoin (BTC), Ether (ETH), and Ripple (XRP) from 2014 through late 2019. A diverse set of predictors was used, including fiat exchange rates, commodities (Gold and Oil), stock market indices (S&P 500 and VIX), interest rates (US 10-year Treasury Yields), and crypto-specific metrics like Hash Rate and Blockchain Size.

Descriptive statistics for the cryptocurrency returns and the financial and crypto-specific predictors used in the study.

Key Insights and Results

Extensive testing revealed that a 4-state NHHM provided the optimal balance between complexity and accuracy. For Bitcoin, these states clearly mapped to Bear (negative returns, high volatility), Bull (positive returns, low volatility), and Calm (near-zero returns, very low volatility) regimes.

A primary finding is that external variables primarily influence the probability of the market changing states rather than just affecting returns directly. For example, the VIX and Treasury Yields were found to be universal predictors of regime shifts across all three coins. Furthermore, the NHHM captured state-dependent effects where a predictor might be significant in a "Bull" state but irrelevant in a "Calm" state.

In terms of performance, the 4-state NHHM outperformed all benchmarks in one-step-ahead forecasting, showing the lowest Mean Squared Forecast Error (MSFE). The model's 1% credibility region captured roughly 95% of observed returns, compared to only 25-28% for the standard Random Walk model.

Conclusion

While the 4-state NHHM offers superior forecasting dominance, the frequent alternation of regimes in cryptocurrency markets presents challenges for long-term persistence. Future work may involve adapting these models for Proof-of-Stake protocols or hybridizing Bayesian HMMs with deep learning architectures to capture more complex temporal dependencies.

Summary updated

Abstract

In this paper, we consider a variety of multi-state Hidden Markov models for predicting and explaining the Bitcoin, Ether and Ripple returns in the presence of state (regime) dynamics. In addition, we examine the effects of several financial, economic and cryptocurrency specific predictors on the cryptocurrency return series. Our results indicate that the Non-Homogeneous Hidden Markov (NHHM) model with four states has the best one-step-ahead forecasting performance among all competing models for all three series. The dominance of the predictive densities over the single regime random walk model relies on the fact that the states capture alternating periods with distinct return characteristics. In particular, the four state NHHM model distinguishes bull, bear and calm regimes for the Bitcoin series, and periods with different profit and risk magnitudes for the Ether and Ripple series. Also, conditionally on the hidden states, it identifies predictors with different linear and non-linear effects on the cryptocurrency returns. These empirical findings provide important insight for portfolio management and policy implementation.
BibTeX
@inproceedings{koki2021exploring,
  author = {Constandina Koki and Stefanos Leonardos and Georgios Piliouras},
  title = {{Exploring the Predictability of Cryptocurrencies via Bayesian Hidden Markov Models}},
  booktitle = {Research},
  year = {2021},
  doi = {10.1016/j.ribaf.2021.101554},
}
Figure