Conditional variance models and causality analysis for the cryptocurrency market: An application

This thesis encompasses an application that examines and tests concepts related to cryptocurrencies, blockchain, and volatility. It touches upon the philosophical and potential dimensions of data security, privacy, and, most crucially, transparency offered by blockchain technology. The study investigates the presence of volatility clustering and leverage effects in cryptocurrency markets which differ from traditional financial markets. Additionally, it tests for statistical causal relationships in the fluctuations of cryptocurrency prices. The study also includes tests suggesting that more advanced volatility modeling might be suitable for predicting the high volatility in cryptocurrency markets. The data used in the study is based on the five highest-volume cryptocurrencies in the markets as of September 2023. In this context, various models of GARCH (Generalised Autoregressive Conditional Heteroskedasticity), such as EGARCH (Exponential GARCH), GJR-GARCH (developed by Glosten, Jagannathan, and Runkle), and TGARCH (Threshold GARCH), have been applied. Distribution characteristics have been considered in assigning the appropriate GARCH model. Assumptions of suitability were determined by skewness, kurtosis, and high log-likelihood values, leading to appropriate p and q-value assignments in GARCH models. Furthermore, the robustness of the established models was tested through the analysis of standardized residuals and Jarque-Bera test analyses. All tests in the study were conducted using Python, R, and Eviews softwares. Excluding stable cryptocurrencies the best autoregressive lag degree (p) = 4 and moving average lag degree (q) = 3 were found for EGARCH, GJR-GARCH, and TGARCH in the trials. For stable cryptocurrencies, these values were determined as (1,1). Therefore, (1,1) models were established in GARCH, EGARCH, GJR GARCH, and T GARCH analyses, compared with the (4,3) model. The most suitable volatility clustering effect was observed in the GJR-GARCH (4,3) model. The most suitable leverage effect was found in the EGARCH (4,3) model. EGARCH and GJR-GARCH models showed a more positive approach in models established with stablecoins and cryptocurrencies. In addition to GARCH model variations, Granger and Yamamoto causality tests were conducted on relevant cryptocurrencies, examining causal relationships among them. The primary reasons for the outstanding suitability of some models of GARCH analysis for specific cryptocurrencies were analyzed through causality relationships. These causality examinations observed mutual influencing potentials between Ethereum and Bitcoin. However, no causality was observed for other stable cryptocurrencies. Keywords: Blockchain, Cryptocurrencies, Conditional Variance, Asymmetric GARCH Models, Granger, Yamamoto, Causality Tests.