April 2017: Data Science in Banking

Data Science Slovenia Meetup series will re-start in 2017! The first meetup in new season will happen on April 19 at 18:00 at NLB's CIP center.

https://secure.meetupstatic.com/photos/event/3/f/f/e/600_460156382.jpeg

The main theme will be how machine learning, statistics and econometrics are applied in banking. We'll host three experts sharing their experience with modeling propensity to buy and one of the core risk management challenges -- likelihood that a borrower will be unable to meet its debt obligations.

1. Credit Rating Models in Banking

Banks use models for almost anything, from propensity scoring, cross-selling, upselling, churn to calculating LGD (Loss Given Default) and PD (Probability of Default). Since the crysis in 2008, banks (and regulators) are working hard to minimize potential credit losses. A big part of that effort is in the hands of credit rating models. Peter Konda, PD modeler at NLB, will talk about challenges traditional banks face when developing scoring models, data sources used and algorithms employed.

About Peter Konda
Peter Konda is originally a software engineer who started as a BI developer, but later found great interest in credit risk modeling using supervised learning on large amounts of rich data. He is currenlty working at NLB d.d. on corporate models for SME.

2. Probability of Default Modeling using Machine Learning

In the presentation we will try to answer the question: What is the probability that the corporate borrower will default in the following year, and what rating grade should it be assigned into (A,B,C or D)? We will discuss the implementation of the rating system for corporate borrowers. The main stages for the model development are: data cleaning and transforming, univariate analysis, variable and model selection, calibration, setting an optimal rating structure, implementation and statistical validation of the estimated probabilities of default. The focus will be on the usefulness of machine learning algorithms and genetic algorithms at different stages of model development. In particular, we will discuss random forests, neural networks and differential evolution algorithms.

About Aljoša Ortl

Aljoša is a credit risk manager at a local bank and a teaching assistant for economics courses. Currently he is responsible for data preparation, analysis and modelling of credit portfolio risk. His main research areas include econometrics, credit risk modelling, financial risk management, and machine learning. With his background in quantitative finance and increased interest in computer science he is trying to combine the areas of econometrics (economics), statistics (applied mathematics) and machine learning (computer science) to solve risk management problems.

3. “Propensity to Buy” modeling with Machine Learning

In the short presentation we will try to answer the question: What is the probability (propensity) that an old customer will buy another product? The main stages for the model development are: data cleaning and transforming, univariate analysis, variable and model selection, calibration, implementation and statistical validation of the estimated probabilities to buy. The focus will be on the usefulness of machine learning algorithms and genetic algorithms at different stages of model development. In particular, we will discuss classification and regression models with use of professional corporate modeling tools.

About Zoran Jurij Tomsits, Ph.D.

Zoran is a researcher with Ph.D. and Austrian economics school of thought background - doing research mostly on the global macroeconomy and business intelligence fields. Also author of various banking risk management, regulatory reporting and marketing solutions; e.g. Strategic and Analytical CRM, Basel III, etc. with use of big data, predictive analytics, and data science methodologies.