Exploring determinants of network stochastic dominance ratios: a causal approach using explainable AI

Various financial ratios are recognised as elements that determine investment decisions, making it essential to identify what factors influence these ratios. The calculation of a ratio is often depicted as a relationship, often in the form of a fraction or percentage, and demonstrates the frequency with which one item is included inside another. The limitations of causal relationships that are derived from observational data are, however, frequently disregarded. We employ structural causal modelling to ascertain the inherent relationship between performance and risk metrics and the network stochastic dominance ratio, as well as how this causal framework influences investment product selection. The network stochastic dominance ratio is an attractive tool for ranking assets with respect to basic stochastic dominance principles. The findings indicate that the extreme Gradient Boosting (XGBoost) technique outperforms the quantile regression method in predicting the network stochastic dominance ratio. To interpret the significance of features, the Shapley Additive Explanations (SHAP) method is employed. The results substantiate the causal importance of network stochastic dominance ratio elements and show the significance of distributional characteristics (Kurtosis) and risk metrics (Max Drawdown and Expected Shortfall) in determining the stochastic dominance ratio. Our research is essential for linking stochastic dominance theories with empirical validation beyond mere correlations.