Early warning systems for cryptocurrency markets: Predicting ‘zombie’ assets using machine learning

The cryptocurrency market harbours a hidden risk: assets that silently disappear from trading, leaving investors stranded. These’zombie’ cryptocurrencies technically exist but become temporarily untradable on exchanges, ranging from weeks to permanent delisting. This study predicts which cryptocurrencies are at risk of becoming zombies using predictors derived from return statistics, trading volume, market capitalisation, and asset-specific features. Our sample includes cryptocurrencies listed for at least 210 days between January 2015 and December 2022. We employ various machine learning algorithms and novel explainable AI (XAI) tools, including permutation-based feature importance and partial dependence plots (PDPs), to identify and analyse key factors influencing zombification risk. Our machine learning models achieve 84% out-of-time balanced accuracy in predicting whether a cryptocurrency will become a zombie within the next 28 days. Tree-based approaches, particularly random forests, significantly outperform traditional econometric methods. Trading volumes and past returns emerge as the most influential predictors.