C-Vine Copulas for Multi-Stock Mean-Reversion Signals
Summary
This document presents a C-vine copula method for statistical arbitrage across a cohort of stocks. It transforms daily returns into uniform pseudo-observations using empirical cumulative distributions, selects a C-vine structure by fitting candidates and comparing AIC, and combines the resulting joint density with numerical integration to estimate a target stock’s conditional probability given the others. Values below or above the midpoint indicate returns unusually low or high relative to the modeled dependence, and can inform a mean-reversion trade.
The discussion emphasizes that structure selection and conditional probabilities depend on the fitted copula families and parameters. It describes alternative ways to position the target stock in the vine and disputes the cited paper’s choice, arguing for a target-centered structure. Candidate selection and dependence modeling can be computationally demanding and vulnerable to overfitting; the document provides a workflow and methodological critique, not evidence of profitability or robust out-of-sample results.
Key ideas
- A C-vine represents multivariate dependence through linked bivariate copulas and marginal distributions.
- Daily stock returns are converted into uniform pseudo-observations using empirical distributions fitted on training data.
- Candidate vine structures can be compared using AIC or likelihood before calculating trading-period densities.
- Numerical integration of the joint density gives a conditional probability used to identify relative mispricing.
- Signals are model-dependent, and structure selection may introduce computational burden and overfitting risk.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.