FATF grey-listing and capital flows : a generalized synthetic control approach

In the presence of time-varying confounding, traditional inference methods such as Difference-in-Differences (DiD) are vulnerable to bias because they rely on fixed additive differences between treated and control units when imputing counterfactual outcomes (Xu, 2017a). This motivates inference methods that implicitly or explicitly model time-varying counfounding. This report examines the performance of the Generalized Synthetic Control (GSC) method studying as an application the effects of grey-listing by the Financial Action Task Force (FATF) on components of gross capital inflows and outflows. This is followed by Carlo simulation evidence that considers the performance of the GSC method against the DiD model and Interactive Fixed Effects (IFE) model by Bai (2009). Limited evidence of listing effects on components of gross capital flows were identified. Model fit concerns, however, reframe this conlusion as a question of limited statistical power rather than evidence that grey-listing generally has no impact on capital flows outcomes. Monte Carlo evidence suggests that the GSC method is a competitive inference method in the presence of lacking common support in factor loadings between treated and control units. When the relevance of factor structure is weak for control units specifically, the GSC method is generally biased. Finally, simulation exercises showed that factor-based estimators, IFE and GSC, incur an estimation penalty from a misspecified data-generating process in a regime characterised by an overall weak factor structure and covariate relevance along with a high magnitude of unstructured idiosyncratic noise. Under this regime, the DiD remains comparatively stable, although generally biased.