Volatility plays a crucial role in trading and risk management. These days, more portfolio managers are aware of and utilize volatility measures, even if they do not trade options. For example, we have previously discussed volatility-managed portfolios, where asset volatilities are used to size positions.
Reference [1] contributes to this body of research by classifying stocks according to their volatilities and then trading them. Specifically, the authors first employ the Gaussian Mixture Model (GMM) to group stocks with similar volatility characteristics. They then apply a causal inference framework based on the Granger Causality Test (GCT) and augmented with Effective Transfer Entropy (ETE) to identify lead-lag relationships, ultimately trading based on market timing signals generated by the lead stock.
The article pointed out,
This paper has put forward a novel methodology for volatility-driven statistical arbitrage in the context of stock market forecasting, which integrates sophisticated statistical techniques with machine learning models. By employing the GMM clustering algorithm to classify stocks according to mid-range volatility, we have demonstrated the potential of such clusters to act as indicators for predicting price movements. This method, combined with a robust causality analysis framework involving Granger Causality Tests, Peter-Clarke Momentary Conditional Independence, and Effective Transfer Entropy, identifies significant predictive relationships between stocks. Furthermore, the integration of DTW and KNN enhances predictive accuracy by aligning and classifying time series data, thereby enabling the anticipation of profitable trading opportunities.
The effectiveness of this integrated approach is demonstrated by the backtesting results. The trading strategies based on identified volatility clusters and causal relationships consistently outperformed the Buy & Hold benchmark across multiple performance metrics, including total returns, Sharpe Ratio, and maximum drawdown. Notably, the volatility-driven strategy yielded substantial returns with controlled risk exposure, as evidenced by the superior Sortino and Calmar ratios in comparison to conventional strategies.
In short, by using GMM to cluster stocks with comparable volatility characteristics and applying causal inference techniques to determine lead-lag pairs, the authors developed a trading strategy that achieved superior risk-adjusted returns.
This is an interesting development. However, we note two limitations:
- The framework involves multiple models, which increases the number of parameters and the risk of overfitting, and
- It appears that only in-sample tests were conducted.
Despite these limitations, there are valuable insights to be gained from this approach.
Let us know what you think in the comments below or in the discussion forum.
References
[1] Ivan Letteri, Statistical Arbitrage Volatility-Driven with Statistics and Machine Learning Models for Stock Market Forecasting, SN COMPUT. SCI. 6, 918 (2025).
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
