Machine Learning Models, Risk Management Current Regulation and Perspectives

Current machine Learning techniques spreading in banking and quantitative finance brings issues to be face by risk management and financial institutions regulation. The former increases models predictive power but at the same time makes models more complex and has to struggle to better explain the role, weight and direction of explanatory variables in the outcomes as well as the sensibility of the result to changes in the feature, different technics have been developed to deal with this problem but there are still important tasks to be completed. On the other side the later has found that the use of internal models as regulated in early Basel regulation stages (Basel II) foster the risk sensitivity for regulatory capital allocation but may lead to undue complexity and so reducing comparability among banks risk. The regulatory response in Basel IV regulation has been to discouraged the use of internal models to reduce the impact of undue complexity in models to improve regulatory capital comparability in the banking system. This is the environment that machine learning models for regulatory purposes have to deal with.