ABSTRACT
The spectacular nature of bitcoin price crashes baffles market spectators and prompts routine warnings from regulators cautioning that cryptocurrencies behave in contra to the fundamental properties that traditionally define what constitutes money. Arguably most concerning to the public is, first, bitcoin's unprecedented price volatility relative to other asset classes and, second, its seemingly detached price behavior relative to time-honored economic and market fundamentals. In an attempt to create an early warning system of bitcoin price crash risk using generalized extreme value (GEV) and logistic regression modeling, this study integrates order flow imbalance, along with several control factors which reflect blockchain activity and network value, in order to nowcast bitcoin's price crashes. From a data analysis perspective, and despite their dissimilar distributional underpinnings, the GEV and logistic models perform comparably. When evaluating the type I and type II errors which these models yield, it is shown that their performance is comparable in terms of accuracy. In addition, it is also shown how the proportion of type I and type II errors can shift dramatically across probability cutoff tolerances. Towards the end of this study, time varying probabilities of a price crash are shown and evaluated. The sample range in this study encompasses the SARS-CoV-2 (Covid-19) time period as well as the recent scandal and collapse of the FTX cryptocurrency exchange.