Deep Learning in Banking

Foreword

by NAEEM SIDDIQI

Lending has not really changed in centuries. Whether someone was borrowing against their future crops in Ancient Mesopotamia, financing trading in the Middle Ages, or getting a P2P microloan via an app—the metrics have been the same. We traditionally have considered the “5Cs” of credit, namely capital, conditions, collateral, character, and capacity to assess the creditworthiness of any borrower. We still look at the exact same broad concepts, but the variables that determine them have constantly evolved. Early bankers likely would have looked at probabilities of good crops or that of ships sinking at the Cape of Good Hope, or perhaps who the parents of the borrowers were in addition to the expected financial information. The advent of credit bureaus created an excellent, centralized, and trustworthy source of information on borrowers’ credit history. This was relatively simple data with a few variables representing traditional credit activities and behavior and has served us well in countries where large portions of the populations have bureau reports. Fast-forward to the digital age, and we are now trying to gauge the same information in a sea of data from every source imaginable: banking transactions, online commerce, social media, cell phone metadata, shopping behavior, driving behavior, car and homeownership, education, and online business ratings to name very few. In addition, these data exist in both structured and unstructured formats, such as ...