LIU Yonghui, YAO Zhao, WANG Jing, LIU Shuangzhe
In the realm of financial data analysis, where data often exhibits volatility clustering, heteroscedasticity, and asymmetry, capturing these inherent features authentically requires the adoption of conditional heteroscedasticity models within a skewed distribution framework. This study addresses the Bayesian statistical diagnostic challenges associated with skew-normal GARCH models. The research begins by employing the Griddy-Gibbs algorithm for effective parameter estimation within the skew-normal distribution embedded in the GARCH model. The investigation considers three distinct sources of perturbations: disruptions from prior assumptions, anomalies within the data, and variations in the model itself. To facilitate comprehensive statistical diagnosis, the study leverages three objective functions: Bayes factor, Kullback-Leibler divergence, and posterior mean, enhancing the precision of the diagnostic process. Empirical validation is achieved through rigorous numerical simulations, conclusively establishing the method's efficacy and resilience. This is further supported by an empirical application involving GARCH modeling for Chevron Stock. By utilizing the skew-normal distribution to encapsulate weekly logarithmic returns, the study empirically underscores the distinct advantages of Bayesian local influence analysis, demonstrating superior results.
