[Paper Review] Correlations of extreme stock returns within a non-Gaussian one-factor model

Pierre Cizeau, Marc Potters+1 more|arXiv (Cornell University)|Jun 2, 2000

Complex Systems and Time Series Analysis5 references5 citations

TL;DR

This paper investigates whether extreme stock return correlations during high volatility periods can be explained by a one-factor model using the market return as the dominant factor. It finds that most correlation dynamics are captured by conditional averages, but subtle effects like Lillo-Mantegna skewness require extending the model to include market-dependent residual variance and skewness.

ABSTRACT

It is commonly believed that the correlations between stock returns increase in high volatility periods. We investigate how much of these correlations can be explained using conditional averages within a simple onefactor description. Using surrogate data with the true market return as the dominant factor, we show that most of these correlations can be accounted for. However, more subtle e#ects (such as the recently discovered LilloMantegna skewness) require an extension of the one factor model, where the variance and skewness of the residuals depend on the market return. 1 Introduction Understanding the relationship between the statistics of individual stock returns and that of the corresponding index is a major issue in several finance problems such as risk management [1] or market micro-structure modeling. It is also crucial for building optimized portfolios containing both index and stocks derivatives [2]. Although the index return is the (weighted) sum of stock returns, it...

Motivation & Objective

• To understand how much of the increased correlation in extreme stock returns during high volatility can be explained by a simple one-factor model.
• To assess whether the market return alone, as the dominant factor, accounts for the observed correlation dynamics in extreme market conditions.
• To investigate whether additional features—such as skewness in residuals—require model extensions beyond standard conditional averages.
• To evaluate the limitations of the one-factor model in capturing subtle statistical dependencies like the Lillo-Mantegna skewness.

Proposed method

• Using surrogate data generated from the true market return to simulate stock returns under a one-factor structure.
• Applying conditional averaging techniques to model stock return correlations based on the market return's realized value.
• Extending the one-factor model to allow the variance and skewness of residuals to depend on the level of the market return.
• Comparing empirical correlation patterns with those generated by the model to assess explanatory power.
• Employing statistical validation to isolate the contribution of market-driven dynamics from higher-order dependencies.
• Analyzing extreme return events to test model performance under high volatility regimes.

Experimental results

Research questions

• RQ1 To what extent can the observed increase in stock return correlations during high volatility be explained by the market return as a one-factor?
• RQ2 Are there residual statistical patterns—such as skewness in return distributions—that the standard one-factor model fails to capture?
• RQ3 Does the Lillo-Mantegna skewness effect persist when conditioning on the market return, and if so, can it be modeled through dynamic residual moments?
• RQ4 How do conditional variance and skewness of residuals influence the correlation structure of extreme stock returns?

Key findings

• The one-factor model with conditional averages explains the majority of the observed increase in stock return correlations during high volatility periods.
• Most correlation dynamics in extreme market conditions are captured by the market return's influence on individual stock returns through conditional expectations.
• The Lillo-Mantegna skewness effect—where negative returns show stronger correlations—cannot be explained by the standard one-factor model alone.
• Extending the model to allow market-dependent variance and skewness in residuals significantly improves its ability to reproduce empirical correlation patterns.
• The model extension successfully captures subtle asymmetries in return dependence that are absent in simpler models.
• Empirical data shows that residual skewness is systematically higher in bear markets, indicating a need for non-Gaussian, state-dependent residual distributions.