Regime identification is important in portfolio and risk management. There are many ways to classify market regimes, for example, based on market direction, such as bullish, bearish, or sideways. Another common classification is based on volatility regimes, such as high or low volatility. Most existing methods for detecting volatility regimes rely on single-point data, such as implied volatility indices or realized volatility measures.
Reference [1] proposes a new regime classification approach based on the entire volatility surface. The method first calculates local gradients, defined as the partial derivatives of implied volatility with respect to moneyness and maturity. These gradient changes are then clustered using an unsupervised algorithm to identify recurring structural transformations of the volatility surface.
The author pointed out,
My goal was to analyze structural changes in the implied volatility surface and study how the surface evolves over time. In this study, I have successfully developed a methodology to represent and quantify daily structural changes in the IV surface using local gradients. I have also used unsupervised clustering algorithm to identify distinct types of surface transformations. I have also included interpretations for some of the clusters in the previous section.
My analysis revealed that there is a number of distinct types of surface transformations that can be identified and interpreted. As expected, the vast majority of daily IV surface changes were classified as noise because structural changes in the volatility surface do not occur often. Therefore, the identified clusters have sizes of up to 18 samples. Notably, several clusters represented specific skew or term structure dynamics for different levels of maturity and moneyness.
In summary, the paper demonstrated that the clusters correspond to specific structural changes in skew and term structure rather than random fluctuations. For example:
- Cluster 1: changes mainly affect short-term maturities (1–2 months) and alter the slope across moneyness, indicating localized movements in the front of the surface.
- Cluster 2: shows term-structure rotation, where longer-maturity implied volatilities fall relative to short maturities for some moneyness levels, making the term structure steeper.
- Cluster 3: reflects a flattening of the skew, where higher-moneyness volatilities decrease relative to lower-moneyness ones, interpreted as increased demand for downside protection (more bearish sentiment).
This paper is exploratory in nature, as it does not prove the economic benefits of using the proposed volatility surface clustering technique. However, it points to an important research direction: volatility regime detection may benefit from analyzing the full volatility surface rather than relying on single volatility indicators, and the paper proposes a practical framework for doing so.
Let us know what you think in the comments below or in the discussion forum.
References
[1] Dzhafarov, Shakhin (2025). Detecting Structural Evolution of Implied Volatility Surface Using Gradient-Based Features: A Machine Learning Approach to Market Regime Detection. Master’s thesis, Aalto University.
Further questions
What's your question? Ask it in the discussion forum
Have an answer to the questions below? Post it here or in the forum
