Nifei Lin,Yingda Song,L. Jeff Hong
Abstract:This paper addresses the estimation of the systemic risk measure known as CoVaR, which quantifies the risk of a financial portfolio conditional on another portfolio being at risk. We identify two principal challenges: conditioning on a zero-probability event and the repricing of portfolios. To tackle these issues, we propose a decoupled approach utilizing smoothing techniques and develop a model-independent theoretical framework grounded in a functional perspective. We demonstrate that the rate of convergence of the decoupled estimator can achieve approximately $O_{\rm P}(\Gamma^{-1/2})$, where $\Gamma$ represents the computational budget. Additionally, we establish the smoothness of the portfolio loss functions, highlighting its crucial role in enhancing sample efficiency. Our numerical results confirm the effectiveness of the decoupled estimators and provide practical insights for the selection of appropriate smoothing techniques.
What problem does this paper attempt to address?
### What problem does this paper attempt to solve?
This paper aims to solve the problem of efficient estimation of systemic risk measure CoVaR (Conditional Value - at - Risk). Specifically, CoVaR quantifies the risk of a financial portfolio when another financial portfolio is in a risky state. The author identifies two main challenges:
1. **Conditioning on zero - probability events (ZPE)**: That is, dealing with the condition {µ(Z) = qα}, which is an event that is almost impossible to directly observe.
2. **Repricing of portfolios**: Since many financial derivatives lack closed - form pricing formulas, they need to be repriced through additional simulation experiments.
To solve these problems, the author proposes a decoupling method and develops a model - independent theoretical framework from a functional perspective. This method uses smoothing techniques to improve sample efficiency and shows that its convergence rate can reach approximately \( O_P(\Gamma^{-1/2}) \), where \(\Gamma\) represents the computational budget.
### Specific problem description
- **Systemic risk**: Financial institutions and markets are interconnected through various channels, such as lending, trading, and derivatives. Fluctuations in risk factors (such as stock prices, commodity prices, and interest rates) may spread through these channels, leading to systemic risk. For example, during the 2007 - 2009 financial crisis, the risks of structured investment vehicles spread to other healthy institutions and ultimately triggered a global crisis. Therefore, understanding and managing systemic risk is crucial for maintaining financial stability.
- **Definition of CoVaR**: CoVaR is defined as the β - Value - at - Risk (VaR) of a portfolio loss, conditional on another portfolio loss being at its α - VaR. Mathematically, CoVaR satisfies the following equation:
\[
\Pr\{\pi(Z) \leq \text{CoVaR}_{\alpha, \beta} \mid \mu(Z) = q_\alpha\} = \beta
\]
where \( q_\alpha \) represents the α - VaR of the portfolio loss \(\mu(Z)\).
- **Challenges**:
- **Zero - probability events (ZPE)**: Dealing with the condition {µ(Z) = qα} is a zero - probability event and cannot be directly observed from the simulation data.
- **Repricing**: When there is no closed - form expression, two - layer simulations must be used for repricing, which increases complexity and computational burden.
### Solution
The author proposes a decoupling method to separate surface fitting from CoVaR estimation in order to improve sample efficiency and simplify asymptotic analysis. The specific steps include:
1. **First stage**: Simulate m outer - layer scenarios and l inner - layer observations (l can be as small as 1) for each outer - layer scenario, and use smoothing techniques to construct function approximations of the portfolio loss functions µ(z) and π(z).
2. **Second stage**: Generate n new outer - layer scenarios and use the function approximations obtained in the first stage to execute a batch - processing estimation algorithm to obtain the CoVaR estimator.
This method not only improves the efficiency of the CoVaR estimator but also enables the number of inner - layer observations to be less than the number of outer - layer scenarios, thereby significantly reducing the computational cost.
### Conclusion
Through this method, the author proves that the proposed decoupled CoVaR estimator can achieve a convergence rate of approximately \( O_P(\Gamma^{-1/2}) \), provided that appropriate smoothing techniques (such as linear regression, kernel ridge regression, or neural networks) are used. This makes CoVaR estimation more robust and efficient in high - dimensional settings.