Christine Jeanneaux

Jury

Abstract

This thesis analyzes herding behaviors and contagion phenomena in cryptocurrency markets. We emphasize the importance of distinguishing between rational and irrational forms of herding. The first chapter reexamines the classical methods used to detect herding, based on cross-sectional return dispersion (CSSD and CSAD). Traditionally, polarization or compression of return dispersion is interpreted as evidence of herding. However, these models do not clearly distinguish blind, irrational imitation from a rational collective reaction driven by public or asset-specific information, or by shifts in investors’ risk aversion. The chapter therefore enhances the original methodology by incorporating proxies for information flows and systemic risk characteristics. The results show that most of the observed reduction in dispersion is explained by rational adjustments to specific signals, challenging the usual behavioral interpretation and suggesting that traditional measures tend to overestimate investor irrationality in cryptocurrency markets. The second chapter proposes an alternative model for detecting herding, better suited to the heterogeneity of the cryptocurrency market. Our model analyzes, for each asset, the gap between realized and expected returns, estimated first via the CAPM and then using the multifactor models of Fama–French and Carhart. This approach captures informational flows more effectively and identifies episodes where crypto-assets simultaneously deviate from their theoretical returns. The results highlight herding behaviors particularly during bearish market conditions, consistent with intensified collective reactions under stress. However, rolling-window robustness tests temper the strength of our findings, showing that herding detection remains sensitive to methodological choices. The third chapter examines contagion among layer-0, layer-1, and layer-2 cryptocurrencies to understand how collective behaviors influence the transmission of shocks. Using the methodology of Forbes and Rigobon (2002) and DCC-GARCH and BEKK-GARCH models, the analysis focuses on three major crises: the invasion of Ukraine, the collapse of Terra Luna, and the bankruptcy of FTX. The results show a strong increase in co-movements during turbulent periods, but no evidence of fundamental contagion: correlations rise without any lasting structural change in the relationships between assets. The observed reactions therefore appear to stem from similar responses to external shocks rather than from contagion mechanisms within the ecosystem’s layers. Overall, this work deepens the understanding of herding and contagion in cryptocurrency markets. It highlights the need for methods tailored to their specific structure and clarifies the distinction between genuinely irrational behaviors and rational adjustments to information.