Do Warnings Change Behavior? Money-laundering, Grey-listing by the FATF, and Cross-border Financial Flows
Date issued
April 2026
Subject
Capital Inflow;
Capital Flow;
Rating;
Gross Domestic Product;
Exchange Rate;
Machine Learning;
Regulation;
Correspondent Bank
JEL code
C14 - Semiparametric and Nonparametric Methods: General;
F21 - International Investment • Long-Term Capital Movements;
F36 - Financial Aspects of Economic Integration;
F38 - International Financial Policy: Financial Transactions Tax; Capital Controls;
G15 - International Financial Markets
Category
Working Papers
Does Financial Action Task Force (FATF) grey-listing disrupt cross-border financial flows? The answer is not obvious. Grey-listing imposes no formal transaction restrictions, yet it can trigger compliance responses by financial institutions facing heightened regulatory risk and the prospect of future blacklisting. Identification is further complicated by anticipation, as firms may adjust during the evaluation process, attenuating observed changes at the time of designation. We combine banking transactions, balance-of-payments capital flows, and foreign-exchange data with complementary identification strategies to isolate the incremental impact of formal grey-listing. To address high-dimensional confounding and separate medium-run adjustments from announcement-driven dynamics, we pair Double Debiased Machine Learning estimates in a fixed-effects panel with a Regression Discontinuity in Time design that exploits the sharp publication timing of FATF decisions. Across approaches, grey-listing produces economically meaningful contractions in inflows: banking and capital inflows fall by 1.3--2.6 percent of quarterly GDP, while outflow responses are smaller and less robust. Regression discontinuity estimates corroborate these results, showing a 1.2--1.4 percentage point drop in inflows at the listing cutoff. Delisting triggers only partial recovery (40--70 percent), consistent with persistent frictions in correspondent banking relationships.
Generative AI enabled