Central Counterparties (CCPs) ensure market stability and guarantee trade settlement during defaults by employing robust risk management frameworks such as margin setting and scenario-based stress testing to ensure sufficient financial safeguards against market shocks. Construction of asset price return scenarios broadly falls into three approaches:
Option contracts have short lifespans, typically maturing within a week, which limits the historical data available for a contract. Additionally, historical price returns of contemporaneous option contracts cannot be clubbed together to differing price surfaces. For these reasons, historical simulation is not feasible for options. Traditionally, CCPs have employed a parametric modelling approach with underlying risk factors such as asset price returns and implied volatility (IV) returns for the generation of option price scenarios. However, option contracts tend to have a sparse IV surface because of liquidity concentration in near term contracts and strike prices in the At-the-Money (ATM) region i.e. strike prices near the price of the underlying asset. Therefore, parametric modelling involves performing certain adjustment to the IV surface such as bucketing or fitting a parametric surface. This leads compression of the IV surface and introduction of biases in resulting scenarios due to sensitivity to model parameters. Moreover, the generated scenarios have inherent modelling assumptions based on the choice of option pricing framework. These issues raise an important question: can we reliably generate option price return scenarios through an alternate approach which addresses the issues with the parametric modelling approach?
In our recent research paper presented at the WFEClear conference, we have sought to develop an alternative approach for scenario generation of option contracts which does not depend on an option pricing framework and involve adjustments to the price surface. The paper provides the detailed description of the approach and evaluation against a widely used parametric modelling approach on data from the Indian option market. We find the proposed approach demonstrates superior performance in generating a forecast of potential price returns.
A risk neutral distribution is a market-implied estimate of future prices of an asset, derived from the observed option prices. Under a risk neutral framework, the asset is expected return on the asset is equal to the risk-free rate. In other words, a daily risk-neutral distribution (RND) summarises how market participants are pricing future uncertainty based on the option prices of that day. The shape of the distribution tells the expected direction of the asset prices. Difference between two risk neutral distributions will reflect the shift in the market expectation about the future price of the asset.
The proposed approach involves estimating the daily risk neutral distributions and computing the daily shift in their values reflecting the change in shape of the distribution. The change in RND values and shape will express different market scenarios, from low volatility regimes to extreme market conditions. We train a generative model to learn and model these distributional change dynamics such that unseen, synthetic values of RND change can be generated and applied to any risk neutral distribution to get simulated versions of the next-day distribution. Option price scenarios at a given strike may be generated by repricing from these distributions. There are a few key advantages to this approach. First, estimation of the risk neutral distribution does not depend on any option pricing framework and works directly with the option prices without the need of any underlying risk factors. Second, the entire price surface is used and does not require any adjustments such as bucketing or fitting a parametric curve. Finally, based on the comparison with the benchmark method, the proposed method has shown better performance consistently across different types of maturities and strike regions. It is also shown that the approach does not demonstrate any regime specific behaviour.
The findings suggest practical implications for policy and risk oversight. Notably, the model-independent approach mitigates reliance on assumptions that may not hold under evolving market conditions, enabling institutions to base assessments on observed price behaviour and enhancing credibility while minimising model bias. The proposed approach may also be extended to incorporate existing market regimes as a conditioning variable. The output price scenarios may be used in common CCP risk measures such as Value at-Risk or Expected Shortfall. Furthermore, the approach aids in model validation by assessing whether current models systematically underestimate or overestimate risk.
Disclaimer: The views expressed in the article are solely those of the author and do not necessarily represent the views of NSE Clearing or NSE.