A bayesian non-parametric approach to dynamic conditional angular correlation model with application to portfolio optimization

Abstract

In financial time series analysis, the dynamic conditional correlation model is the most popular method for estimating the conditional covariance matrix, which represents financial risk and is critical for risk management, portfolio optimization, and asset pricing. Traditional covariance matrix estimation is often constrained by the rigid parameter settings and the assumption of the normal distribution, leading to the estimation biases when the markets are not normally distributed. To address these limitations, this paper proposes a Bayesian Non-parametric Dynamic Conditional Angular Correlation model based on the Fractionally Integrated GARCH model (BNDCAC-FIGARCH) that incorporates the asymmetric parameter and the student’s t-distribution to increase the adaptability and flexibility. Simulation experiments demonstrate that under overall correlation paths shaped as the sine or ramp functions, our model provides more accurate estimates, showcasing its effectiveness and stability. Empirical studies apply real stock market data, which includes DAX 40, FTSE 100, SSE 50, and CSI 100, to construct the portfolio optimization. The results demonstrate the superiority of the proposed model in terms of both portfolio returns and the reduction of parameter uncertainty. Furthermore, the results indicate that CSI 100 exhibits the weaker asymmetry compared to the other indices, likely due to its higher liquidity and a more accurate reflection of improved economic conditions resulting from national policies.