Bayesian Filtering in a Latent Space to Predict Bank Net Income from Acquiring

摘要

A macro stress-testing and scenario analysis is an important part of an official assessment of any bank regarding its safeguarding and stability. There is a lack of efficient tools for scenario analysis to model uncertainty of the bank financial indicators depending on the main macro-economic parameters. In this work we present a new model for prediction of the bank financial indicators. We develop an approach to filtering in a latent space capable of modeling dependence of a huge cross-section of the indicators on the set of macro-economic parameters. We demonstrate a superior ability of our model to predict bank net income from acquiring compared to standard predictive models.