Measuring Fraud in Banking and its Impact on the Economy: A Quasi-Natural Experiment

This paper suggests a novel approach to measuring fraud in banking and to evaluating its crosssectional and aggregate implications. I explore unique evidence of declining regulatory forbearance from the Russian banking system in the 2010s, when the central bank forcibly closed roughly twothirds of all operating banks for fraudulent activities. I first introduce an empirical model of the regulatory decision rule that determines whether a regulator is likely to run an unscheduled onsite inspection of a suspicious bank in the near future. I estimate the model using unique data on asset losses hidden by commercial banks and discovered by the Central Bank of Russia during unscheduled on-site inspections in the last two decades. I find that the average size of hidden asset losses detected by the rule equals 38% of the total assets of not-yet-closed fraudulent banks, and that the likelihood of fraud detection soared by a factor of 5 after 2013. With quarter-by-quarter predictions from the estimated rule, I form a “treatment” group of likely-to-be-inspected banks and then run a “fuzzy” difference-in-differences (FDID) regression to estimate the effects of the tightened regulation. FDID estimates show that likely-to-be-inspected banks substantially reduced credit to households and firms after the policy started in 2013, compared to similar untreated banks. Interpreting the FDID estimates of credit contraction as a credit supply shock and evaluating the macroeconomic implications of this shock using a VAR model of the Russian economy, I find that Russia’s GDP could have been larger by 7.3% cumulatively by the end of 2016 in the absence of the policy. This is the price the economy pays for reducing fraud in the banking system.