Automation of the Approach to Replicating Data When the Control Group Is Depleted in The Difference-In-Differences Method: Application to IRB Implementation Data Samples

Abstract

Internal-Ratings Based (IRB) approach is one of the founding blocks in the modern credit risk management and regulation. Its implementations by the banks world-wide incentivizes researchers, central bankers and investors to evaluate the outcome versus the non-IRB banks, i.e., the treatment effect. However, there are obstacles in such evaluation. From one side, all the banks may transit to IRB for a relatively modest economy (like Greece). From another side, there is no single IRB transition point (like in the EU where the transition is voluntarily; a single point exists for the USA where the transition was mandatory for the largest banks). That is why we discuss the problem of the ‘control’ group depletion (attrition) in the Difference-in-Differences method. We provide two ways to replicate data using python language. The first code is based on the matrix structure (object * time), the second – panel data structure. Comparing both ways of the data replication (resampling) we choose the second algorithm due to it versality and faster computing results, than the first one.