A Causal Analysis of Market Contagion: A Double Machine Learning Approach

Making reliable causal inferences is integral to both explaining past events and forecasting the future. Although there are various theories of economic causality, there has not yet been a wide adoption of machine learning techniques for causal inference within finance. One recently developed framework, double machine learning, is an approach to causal inference that is specifically designed to correct for bias in statistical analysis. In doing so, it allows for a more precise evaluation of treatment effects in the presence of confounders. In this article, the author uses double machine learning to study market contagion. He considers the treatment variable to be the weekly return of the S&P 500 Index below a specific threshold and the outcome to be the weekly return in a single major non-US market. In analyzing each non-US market, the other non-US markets under consideration are used as confounders. The author presents two case studies. In the first, outcomes are observed in the same week as the treatment is observed and, in the second, in the week after. His results show that, in the first case study, sizable and statistically significant contagion effects are observed but somewhat diluted due to the presence of confounders. In contrast, in the second case study, more ambiguous contagion effects are observed and the level of statistical significance is measurably lower than those observed in the first case study, indicating that contagion effects are most clearly transmitted in the same week that the dislocation in the S&P 500 occurs.