Testing Granger Non-Causality in Panels with Cross-Sectional Dependencies

Within the last decade, there has been growing awareness that causal inference can improve scientific research in many disciplines as interpretability and robustness become increasingly important (Doshi-Velez and Kim, 2017; Roscher et al., 2020; Marcinkevičs and Vogt, 2020; Moraffah et al., 2020). Causality is a crucial factor for gaining insights into the decision process of algorithms, which has many use cases such as avoiding bias and discrimination (Mehrabi et al., 2019), improving user experience (Zhou and Fu, 2007) and gathering biological insights (Angermueller et al., 2016). If the causal relation between variables is known, causality can be used to study the interaction between statistical units such as estimating the average effect of treatments (Imbens and Rubin, 2015; Holland, 1986), analyze their mediation (Berzuini et al., 2012), detect the root causes of anomalies (Janzing et al., 2019) or quantifying the causal influence of variables in a system (Janzing et al., 2013, 2020).