Abstract
Based on Copula function and GARCH model, this paper establishes the Copula-GARCH-GED model to analyze the portfolio risk of four stocks in different industries—Minsheng Investment, Huayi Brothers, Renhe Pharmaceutical, and Yanghe Shares in the Shenzhen stock market. Using Monte Carlo simulation, the study calculates optimal investment ratios and portfolio Value at Risk (VaR) across various confidence levels.
Introduction: Beyond the Normality Assumption
Traditional financial risk management often relies on the assumption that stock returns follow a normal distribution. However, empirical evidence frequently contradicts this "bell curve" ideal. Real-world markets are characterized by "fat tails"—where extreme events occur more frequently than predicted—and volatility clustering, where periods of high or low volatility tend to persist.
This paper addresses the challenge of accurately estimating Value at Risk (VaR), the maximum potential loss over a given timeframe. By integrating GARCH models to capture individual asset volatility and Copula functions to model complex inter-asset dependencies, the study provides a more robust framework for portfolio risk analysis in the Shenzhen stock market.
Methodology and Data
The study utilizes daily closing prices from January 1, 2014, to December 31, 2015, for four diverse assets: Minsheng Investment (MI), Huayi Brothers (HB), Renhe Pharmaceutical (RP), and Yanghe Shares (YS). This period yielded 650 log-yield data points per stock.
Modeling Volatility with GARCH-GED
To account for the "spiky" nature of financial data, the researchers employed a GARCH(1,1) model. Unlike standard models, they utilized the Generalized Error Distribution (GED). Statistical comparisons using the Akaike Information Criterion (AIC) demonstrated that the GED outperformed the traditional t-distribution in describing these specific stock fluctuations.
Linking Assets with Copula Functions
While GARCH models individual stocks, the Gaussian-Copula function links them to model the joint distribution of the entire portfolio. This approach captures the dependency structure between assets regardless of their individual marginal distributions.
Statistical Analysis and Insights
A critical first step was verifying the statistical properties of the returns. The Jarque-Bera test confirmed that all four stocks exhibited kurtosis values significantly higher than 3 (reaching as high as 75 for some assets), reinforcing the "normality myth."
The Gaussian-Copula analysis revealed relatively low correlation coefficients between the stocks, ranging from -0.19 to 0.24. These low correlations suggest that a portfolio combining these specific assets offers significant diversification benefits, effectively reducing overall risk.
Results: VaR and Portfolio Optimization
Using Monte Carlo simulations, the researchers generated thousands of scenarios to determine the "minimum risk" portfolio weights and corresponding VaR at confidence levels from 90% to 99%.
- VaR Findings: The minimum VaR values ranged from -0.0374 (at 90% confidence) to -0.0649 (at 99% confidence).
- Dynamic Weighting: The optimal asset allocation shifted as the required confidence level increased. Notably, at the 99% confidence level, the model suggested a 100% allocation to Renhe Pharmaceutical to minimize extreme tail risk.
- Model Validation: The model's accuracy was confirmed through the K-S test (p-values > 0.44) and the Kupiec test, which showed that actual market "failures" (losses exceeding VaR) fell within acceptable statistical bounds.
Conclusion and Future Directions
The Copula-GARCH-GED framework proves to be a reliable tool for predicting portfolio risk in the Shenzhen market. However, the authors note that future research could be enhanced by incorporating EGARCH models to account for the "leverage effect"—where negative news impacts volatility more severely than positive news. Additionally, exploring more advanced Copula estimation algorithms could further refine risk precision in increasingly volatile global markets.
Summary updated