Regime-Dependent Predictability of Cryptocurrency Distress: Nonlinear Evidence from Cross-Exchange Dynamics

The study of regime-dependent predictability of cryptocurrency market distress is examined in this paper using the dynamics of cross-exchange spread signals as an innovative early-warning tool. The research offers a three-stage econometric framework comprising: a Markov regime-switching VAR to infer endogenous bull, transition, and bear regimes in the BTC, ETH, BNB, and SOL markets; a Panel Smooth Transition Regression (PSTR) model for the regime-dependent spread-distress predictive relationship; and quantile regression providing the complete profile of the distress distribution. The full sample spans January 2020–December 2025 (2191 daily observations); the January 2020–December 2022 sub-period (1095 observations) is used for baseline in-sample estimation, while January 2023–December 2025 is reserved for out-of-sample validation. The results show that cross-exchange spread dynamics contribute to the occurrence of distress regardless of the market state and confirm that bear markets predict distress better than bull markets (with AUROC values changing from 0.848 to 0.963, respectively). In addition, the results of the PSTR function suggest that the predictive role of the spread increases in stressful periods (from 0.152 to 0.588), as supported by LM tests yielding results with p-values less than 0.001. The suggested model outperforms seven competing models, including MS-VAR, GARCH-M, linear logit, LSTM, and XGBoost, on all evaluation measures (AUROC, Brier score, Diebold–Mariano) across every regime. The results remain stable across different exchange-pair configurations and COVID-19-related sub-samples, and when varying the threshold definition.