Cryptocurrency markets are characterized by large price fluctuations and recurrent episodes of market stress, making the assessment of tail risk particularly important. Standard Gaussian-based models often underestimate the probability of extreme events because cryptocurrency returns exhibit heavy tails and volatility clustering. This thesis investigates the tail behavior of four cryptocurrencies: Bitcoin (BTC), Ethereum (ETH), Tether (USDT) and USD Coin (USDC), using Extreme Value Theory (EVT) to compare the risk characteristics of highly volatile assets and stablecoins within a unified framework. Return series were filtered using GARCH-type models, with ARMA-GARCH specifications adopted for the stablecoins in order to satisfy the assumptions required by EVT. The resulting standardized residuals were analyzed using two univariate EVT approaches: the Block Maxima Method (BMM), based on the Generalized Extreme Value (GEV) distribution, and the Peaks Over Threshold (POT) method, based on the Generalized Pareto Distribution (GPD). Return levels were estimated over a five-year horizon and used as the primary measure of tail risk. The results reveal substantial differences between volatile cryptocurrencies and stablecoins. Bitcoin and Ethereum exhibit considerably larger extreme gains and losses than USDT and USDC. While the stablecoins display Fréchet-type tail behavior, the filtered residuals of Bitcoin and Ethereum do not provide sufficient evidence to reject a Gumbel-type specification. Return level estimates, in line with the nature of the digital assets analyzed, indicate much greater exposure to extreme market movements for BTC and ETH than for the stablecoins. The analysis is subsequently extended to a bivariate EVT framework to investigate extremal dependence between asset pairs. Componentwise Maxima and threshold exceedance models are estimated using both two-stage and full likelihood approaches. Significant extremal dependence is consistently identified between Bitcoin and Ethereum across these estimation methods. Evidence of dependence between USDT and USDC is weaker and not robust across specifications, while pairs combining a stablecoin with a volatile cryptocurrency show no meaningful same-tail dependence. An analysis of asymmetric tail dependence provides limited but suggestive evidence consistent with the safe haven hypothesis, at times attributed to stablecoins during periods of market stress. Overall, the findings demonstrate that EVT provides a useful framework for quantifying tail risk in cryptocurrency markets and highlight substantial differences in both marginal tail behavior and extremal dependence between volatile cryptocurrencies and stablecoins.
I mercati delle criptovalute sono caratterizzati da ampie fluttuazioni dei prezzi e da ricorrenti episodi di stress di mercato, rendendo la valutazione del rischio di coda particolarmente importante. I modelli standard basati sull'ipotesi gaussiana tendono a sottostimare la probabilità di eventi estremi, poiché i rendimenti delle criptovalute presentano code pesanti e volatilità clusterizzata. Questa tesi analizza il comportamento di coda di quattro criptovalute: Bitcoin (BTC), Ethereum (ETH), Tether (USDT) e USD Coin (USDC), tramite la Teoria dei Valori Estremi (EVT), con l'obiettivo di confrontare le caratteristiche di rischio di asset ad alta volatilità e stablecoin all'interno di un framework unificato. Per soddisfare le ipotesi richieste dalla EVT, le serie dei rendimenti sono state filtrate tramite modelli di tipo GARCH, con specificazioni ARMA-GARCH adottate per le stablecoin. I residui standardizzati risultanti sono stati analizzati mediante due approcci appartenenti all’EVT univariata: il metodo dei Massimi a Blocchi (Block Maxima Method, BMM), basato sulla distribuzione dei Valori Estremi Generalizzata (GEV), e il metodo dei Picchi oltre Soglia (Peaks Over Threshold, POT), basato sulla distribuzione di Pareto Generalizzata (GPD). I livelli di rendimento estremo sono stati stimati su un orizzonte di cinque anni e utilizzati come misura principale del rischio di coda. I risultati evidenziano differenze sostanziali tra le criptovalute volatili e le stablecoin. Bitcoin ed Ethereum mostrano guadagni e perdite estreme considerevolmente maggiori rispetto a USDT e USDC. Mentre le stablecoin presentano un comportamento di coda di tipo Fréchet, i residui filtrati di Bitcoin ed Ethereum non forniscono evidenza sufficiente a rifiutare una specificazione di tipo Gumbel. Le stime dei livelli di rendimento estremo indicano, coerentemente con la natura degli asset digitali analizzati, una esposizione ai movimenti estremi di mercato molto maggiore per BTC ed ETH rispetto alle stablecoin. L'analisi viene successivamente estesa a un framework EVT bivariato per indagare la dipendenza estrema tra coppie di asset. I modelli basati sui Massimi Componentwise e sul superamento di soglia sono stimati sia tramite approcci a due stadi (Two-Stage) sia tramite massima verosimiglianza completa (Full Likelihood). Tra questi metodi di stima, una dipendenza estrema significativa è identificata in modo consistente tra Bitcoin ed Ethereum. L'evidenza di dipendenza tra USDT e USDC risulta invece più debole e non robusta al variare delle specificazioni, mentre le coppie che combinano una stablecoin con una criptovaluta volatile non mostrano alcuna dipendenza rilevante nella stessa coda. Un'analisi della dipendenza di coda asimmetrica fornisce evidenza limitata ma suggestiva, compatibile con l'ipotesi di safe haven spesso attribuita alle stablecoin nei periodi di stress di mercato. Nel complesso, i risultati dimostrano che la EVT costituisce un framework utile per quantificare il rischio di coda nei mercati delle criptovalute, e mettono in luce differenze sostanziali sia nel comportamento marginale di coda sia nella dipendenza estrema tra criptovalute volatili e stablecoin.
Modellazione di eventi estremi nei mercati delle criptovalute e degli stablecoin tramite la teoria dei valori estremi univariata e bivariata
ADAMI, LISA
2025/2026
Abstract
Cryptocurrency markets are characterized by large price fluctuations and recurrent episodes of market stress, making the assessment of tail risk particularly important. Standard Gaussian-based models often underestimate the probability of extreme events because cryptocurrency returns exhibit heavy tails and volatility clustering. This thesis investigates the tail behavior of four cryptocurrencies: Bitcoin (BTC), Ethereum (ETH), Tether (USDT) and USD Coin (USDC), using Extreme Value Theory (EVT) to compare the risk characteristics of highly volatile assets and stablecoins within a unified framework. Return series were filtered using GARCH-type models, with ARMA-GARCH specifications adopted for the stablecoins in order to satisfy the assumptions required by EVT. The resulting standardized residuals were analyzed using two univariate EVT approaches: the Block Maxima Method (BMM), based on the Generalized Extreme Value (GEV) distribution, and the Peaks Over Threshold (POT) method, based on the Generalized Pareto Distribution (GPD). Return levels were estimated over a five-year horizon and used as the primary measure of tail risk. The results reveal substantial differences between volatile cryptocurrencies and stablecoins. Bitcoin and Ethereum exhibit considerably larger extreme gains and losses than USDT and USDC. While the stablecoins display Fréchet-type tail behavior, the filtered residuals of Bitcoin and Ethereum do not provide sufficient evidence to reject a Gumbel-type specification. Return level estimates, in line with the nature of the digital assets analyzed, indicate much greater exposure to extreme market movements for BTC and ETH than for the stablecoins. The analysis is subsequently extended to a bivariate EVT framework to investigate extremal dependence between asset pairs. Componentwise Maxima and threshold exceedance models are estimated using both two-stage and full likelihood approaches. Significant extremal dependence is consistently identified between Bitcoin and Ethereum across these estimation methods. Evidence of dependence between USDT and USDC is weaker and not robust across specifications, while pairs combining a stablecoin with a volatile cryptocurrency show no meaningful same-tail dependence. An analysis of asymmetric tail dependence provides limited but suggestive evidence consistent with the safe haven hypothesis, at times attributed to stablecoins during periods of market stress. Overall, the findings demonstrate that EVT provides a useful framework for quantifying tail risk in cryptocurrency markets and highlight substantial differences in both marginal tail behavior and extremal dependence between volatile cryptocurrencies and stablecoins.| File | Dimensione | Formato | |
|---|---|---|---|
|
ADAMI_LISA_905259.pdf
accesso aperto
Dimensione
2.4 MB
Formato
Adobe PDF
|
2.4 MB | Adobe PDF | Visualizza/Apri |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14247/29361