Enhancing SME Digital Lending in Egypt through Credit Scoring Masterclass

The first module provides a general introduction into statistical scoring and machine learning. Moreover, it ensures that the students are/get familiar with the widely applied expected loss metrics, as a standard measure for credit risk. We want to emphasize that scoring is a process, not a secret formula. The success of a scoring system, as any predictive model in banking, relies not only on the technical and analytical capacity of developing staff but also on wide staff ownership, an appropriate customer view and underlying data. The second session highlights the most important success factors on a glance and provides some anecdotal evidence why in some cases models do not work as expected.

For the case studies of this series, general Excel skills are required. In the third session we an optional training on the most required Excel functions, in particular Pivot Tables, LOOKUPs resp. INDEX/MATCH, SUMIFS/COUNTIFS, Filters, Sorting, Conditional formatting, Charts, and the Data Analysis Tool Kit. It is only recommended for participants who feel serious uncertainty in many of those.

This module is the core of the course. Participants will learn everything about developing a credit scoring model on their own. The first part will lay the necessary mathematical and statistical foundations for applied popular machine learning algorithms. It will demystify the statistics and provide a very solid under-standing of what these algorithms do and how they should be interpreted without using “scary math”, Greeks and integrals.

The second part is dedicated to a large case study. We will show, explain, and ultimately teach all necessary work steps from the initial data to the final model – all in MS Excel together with a little plugin. Every individual work step will base on the theoretical foundation laid in the session described above. After that you should be able to apply various machine learning algorithms on your own data, hence train and test your own models.

This module has a closer view on Egyptian regulatory rules with regards to scoring and rating schemes and also presents and dis-cusses general taxonomies and frameworks (e.g. BASEL, IFRS 9). Moreover, it compares respective guidelines from other jurisdictions and aims to derive recommendations for a revised Egyptian framework. This module is partly planned to be an interactive workshop with the regulator group.

In the light of serious climatic effects it is seen as a duty for many banks to anchor respective commitments in their corporate identity and daily procedures. There are increasing efforts, often supported by the international donor community, to facility and accelerate such developments. This module will discuss and present possible approaches and lessons learnt from that journey – in particular how risk-evaluating scoring schemes can be enriched with ESG data to obtain a rating about a customer’s environmental foot-print.

Adequate, accurate and complete data is the most decisive single resource to train predictive models. This module will focus on data importance, common accuracy and consistency issues and typical quick wins by improving a data architecture. Besides, it will link to legal requirements in the Egyptian jurisdiction.

Model development is one thing, implementation the other. This module focuses on next steps to ensure a successful implementation, fully utilizing efficiency gains and at the same time not creating discomfort or fear among staff and clients. It further touches on best-practice process and policy adjustments and provides guidelines for successful system implementation.

Customers expect high service quality, convenience and quick response. Digital financial services are commonly understood as a solution to achieve all of these goals simultaneously. In lending, a fully digitalized workflow from loan application until disbursement can be highly efficient, cut costs and become a strong argument for customers. This session will elaborate on the latest digital lending approaches and how digital scorecards play a role in that. It will also create a link to the previous sessions on effective data management to make the most use of your data.

Many financial services providers target small and medium agricultural businesses, which typically have different risk profiles and extraordinary cash cycles. The industry has made significant attempts and progress in embedding agricultural indicators into credit scoring schemes. This module will give an overview about such techniques and present necessary work steps to develop internal agri scorecards.

Who is interested into more of the mathematical foundations for algorithms presented in the previous module may wish to attend the first two sessions of this fully optional module. Here, unfortunately, some “scary math” cannot always be avoided, but will be used at the bare minimum.

This course is provided in MS Excel to be open for practitioners, who are not familiar with statistical programs and programming languages. However, depending on the amount of data used to train a model, the calculations, though mathematically simple, can become computationally very expensive. In such cases, Excel would not be the most appropriate modelling environment. It is beyond the scope of this course to provide comprehensive guidance for many available ML programs/languages.. Though this cannot be shown in depth in this course, the sessions introduce approaches for Logistic Regression and Random Forest Classifiers in R.