Deep Learning in Banking

Chapter 3Time Series and Recurrent Models

Summary

After reading this chapter, and going through the exercises, you will be able to:

• Understand how banking time series data can be analyzed with deep learning methods.
• Understand the theory and main components behind Recurrent Neural Networks and Temporal Convolutional Networks.
• Train and compare recurrent models for risk modeling.

Time series data is one of the most readily available type of data in banking. Any history of transactions will, by its very nature, generate a time series when segmented and grouped by any given time interval. In corporate lending, companies produce quarterly reports, monthly bond and loan repayment data, and, if they have a credit line and/or a bank account with the institution, high-frequency transactional data from bank accounts and credit lines. In retail banking, the situation is similar. Active customers pay their mortgages, auto and consumer loans monthly or biweekly, pay (or not) their credit card statements partially, or in full, every month. When transacting through their bank accounts, they generate records that show that they receive their salaries, pay bills, and transfer funds to businesses, friends, and family. And there is also a huge secondary source of data in the corporate world if we include stock, bond, and option market trades to which a bank gets access. Such data, while a whole world in and of itself, escapes the core objective of this book. There are numerous applications ...