TL;DR
This paper tackles the challenge of accurately assessing credit risk in China's real estate industry by proposing an enhanced KMV model. By integrating GARCH(1,1) for more robust volatility estimation and a Genetic Algorithm (GA) to optimize the default point, the authors achieve a more precise credit risk evaluation, particularly for distinguishing between distressed and healthy real estate companies.
What Problem Does This Paper Address?
The real estate industry, characterized by high capital requirements and long investment cycles, is inherently susceptible to credit risk. This risk is amplified by the complex interplay of economic cycles, regulatory policies, and the sector's significant contribution to the overall financial system. Traditional credit risk models, like the standard KMV model, often struggle to accurately capture the nuances of China's capital market due to differences in market structure, data availability, and regulatory environments. Specifically, the standard KMV model's assumptions about normal distribution of equity value volatility and its fixed default point parameters may not be optimal for the unique characteristics of Chinese real estate firms, leading to potential inaccuracies in risk assessment and early warning systems.
Methods
The core of this paper is the development of an improved KMV model, termed the GA-GARCH-KMV model. This enhancement involves two key components:
- GARCH(1,1) for Equity Value Volatility: The authors replace the standard KMV assumption of normally distributed equity value volatility with the GARCH(1,1) model. GARCH (Generalized Autoregressive Conditional Heteroskedasticity) is well-suited for capturing the time-varying and clustered volatility often observed in financial markets, making it more appropriate for the volatile nature of stock prices in the real estate sector.
- Genetic Algorithm (GA) for Default Point Optimization: The traditional KMV model uses a fixed formula for the default point (DP = 1*STD + 0.5*LTD). This paper employs a Genetic Algorithm to dynamically optimize these default point parameters (specifically, the coefficients for short-term and long-term liabilities). GA is a powerful optimization technique inspired by natural selection, capable of searching for global optima in complex problem spaces, thus finding more tailored default point parameters for Chinese real estate companies.
The model construction involves calculating asset value and volatility, then using GA to optimize default point parameters, and finally computing the default distance (DD).
Datasets
The study utilizes data from 44 Chinese A-share listed real estate companies. The sample includes both non-ST (no trading suspension) and ST* (previously suspended but currently trading) companies, selected to ensure a balanced representation. Data spans from 2016 to 2020, sourced from the CSMAR Financial Terminal database.
Key Insights and Ideas
- GARCH is Crucial for Real Estate Volatility: The assumption of normal distribution for equity value volatility in standard KMV is insufficient for the real estate sector. GARCH(1,1) provides a more realistic representation of the time-varying and volatile nature of asset prices.
- GA Tailors Default Points: A one-size-fits-all default point formula is suboptimal for China's real estate market. GA effectively tailors these parameters, leading to a more accurate assessment of default risk specific to the industry's characteristics.
- Distinguishing ST vs. Non-ST is Enhanced: The modified model significantly improves the ability to differentiate credit risk levels between distressed (ST) and healthy (non-ST) real estate companies.
- China's Real Estate Market Has Unique Default Drivers: The paper suggests that the optimized default point for Chinese real estate companies is higher than the standard KMV default point, with a greater emphasis on long-term liabilities. This reflects the industry's capital-intensive nature, long development cycles, and potentially government support, which influences how companies manage their debt obligations.
Key Results
The empirical results demonstrate the superiority of the GA-GARCH-KMV model:
- Improved Accuracy: The modified KMV model achieves a higher assessment accuracy rate. For the ST group, accuracy reaches 81.8%, and for the healthy group, it reaches 90.9%. This is a significant improvement compared to the traditional KMV model, which showed 0% accuracy for the default group and 84.8% for the healthy group (considering actual 2021 defaults).
- Meaningful Default Distance Differences: Before correction, the mean default distance for ST and non-ST samples were 1.5680 and 2.2773, respectively. After correction, these values become -3.3171 for ST and 1.8250 for non-ST, highlighting a more pronounced and statistically significant separation between the two groups.
- Optimized Default Point Formula: The GA optimization yielded a new default point formula for Chinese real estate companies: DPT = 1.013*STD + 1.834*LTD. This formula places a higher weight on long-term liabilities compared to the traditional KMV formula.
Limitations and Future Work
- Limitations of Traditional KMV: The paper acknowledges that the traditional KMV model has limitations and its original parameters are not fully applicable to Chinese stock markets.
- Sample Specificity: The study focuses on real estate companies. Future work could explore the applicability of this enhanced model to other industries.
- Data Scope: While the study uses data from 2016-2020, incorporating more recent data, especially post-pandemic, could provide further insights.
- External Factors: The paper touches upon economic recovery and government recognition of real estate as a "pillar industry." Future research could more explicitly incorporate macroeconomic indicators and policy changes into the model.
Summary updated