Artikli pealkiri: Crash Risk Premium in Cryptocurrencies Market through Machine Learning

Autorid: 
Tanveer Ahmad (TalTech)
Tõnn Talpsepp (TalTech)
Syed Jawad Hussain Shahzad (University of Waikato, NZ)

Lühikokkuvõte: 

This paper provides the first comprehensive analysis of the crash risk premium in cryptocurrency markets. Using daily data for more than 2,000 cryptocurrencies from 2015 to 2025, we document that crash risk is pervasive and systematically priced in crypto returns. Portfolios sorted on conventional crash risk measures exhibit economically large and statistically significant return spreads that persist after controlling for standard cryptocurrency risk factors. To understand the drivers of crash risk, we employ a broad set of machine learning models to predict crash risk using information from prices, trading volume, and market capitalization. Nonlinear and ensemble-based models substantially outperform linear benchmarks in out-of-sample prediction, indicating that crash risk is driven by complex and nonlinear dynamics. We further show that machine learning–predicted crash risk has strong economic value: portfolios formed on predicted crash risk earn large and persistent risk-adjusted returns, with high-minus-low portfolios generating significant return. Variable importance and portfolio characteristic analyses reveal that extreme return realizations, volatility, liquidity conditions, and behavioral factors are the dominant determinants of crash risk. Extensive robustness tests across more than 5,000 alternative research designs confirm that the crash risk premium is stable and not driven by specific empirical choices. Overall, our findings establish crash risk as a priced and predictable source of risk in cryptocurrency markets and demonstrate the value of machine learning for identifying and pricing tail risk.

Tallinna Tehnikaülikooli majandusanalüüsi ja rahanduse instituudi avatud teadusseminarid toimuvad tavaliselt kuu teisel ja neljandal kolmapäeval nii kohapealse kui ka veebis osalemisvõimalusega, kui ei ole teatatud teisiti. Ettekanne kestab u 45 minutit, millele järgneb veerand tundi arutelu. Seminar toimub inglise keeles. Ettekande aluseks olev artikkel on üldjuhul kättesaadav seminaril kohapeal. Küsimuste korral võib pöörduda seminaride koordinaatori Karsten Staehr karsten.staehr@taltech.ee poole.