Count Scoring de la Fer or a study on credit scoring as part of broadening one's horizons. Part 1

AntipovSN and MihhaCF

UPD part two here

Part one, in which the Count has not yet become Athos, has not met Milady and everything is fine with him

Introduction from the authors:

Good day! Today we are starting a series of articles devoted to scoring and the use of graph theory in it (T.G.). I hope we have enough fuse, strength and patience, because the topic is quite voluminous and, in our opinion, interesting.

Despite the comic title, we will try to touch upon not comic topics that already affect the lives of many of us, and in the near future they can affect everyone, without exception.

All comic allegories, inserts, etc. are designed to slightly relieve the narrative and not allow it to fall into a tedious lecture. We apologize to everyone who doesn’t get into our humor

Now to the point.

The purpose of this article: in no more than 30 minutes, introduce the reader to the research problem, determine the level of consideration of the problem, describe the basic concept of the study and introduce basic terms.

Terms and Definitions:

A complete graph might look like this:

Aramis was always cunning ... on his mind, even Athos owed him. Porthos, until he met Madame Koknar, could not afford to buy a dressing and managed to owe a beggar to D'artanyan, although, frankly, they mutilated something all the way together ...

Graphs consist of nodes and edges. A node can be directly connected to several other nodes. These nodes are called neighbors.

Well, sort of, with the most basic concepts figured out, you can get closer to the point.

Scoring can be used to evaluate almost anything, which can be expressed in statistical indicators. This is an assessment of the creditworthiness of an individual / legal entity (scoring of the applicant), and an assessment of the likelihood of fraud (scoring from fraud), and an assessment of the insured (insurance scoring), an assessment of the supplier / customer (scoring of the counterparty), assessment of consumer behavior (behavioral scoring), social assessment ("Chinese" scoring), etc.

Graph theory, in turn, is also a universal tool that can be used in any field of activity in which it is necessary to process large, multi-level volumes of data.

These two tools are created for each other, like D'artanyan and Constance ( you just need to monitor Constance normally and not let any Miladya go ).

We will not write anything about the importance and topicality of scoring, for it is enough to take a closer look around and it will immediately become clear that we have been explicitly or not explicitly scoring for a long time, then it will only be more fun.

In the series of articles, we will try to clearly demonstrate how scoring works using graph theory in the banking sector. That is, we will determine the creditworthiness of legal entities (maybe we’ll even hook physicists) based on the data they provide and the relationships they have with other organizations - the so-called “borrower scoring” .

As follows from the official definition, the scoring of the borrower is designed to eliminate the subjectivity of the decision of the credit inspector, reduce the level of internal fraud and increase the speed of decision-making on the loan. Let's see if this is so, expand the candy, so to speak, and see what it is made of.

The banking sector was not chosen by chance - banks have extensive sources of information and are scoring using automation, more and more actively.

A little closer to the point. Remember how D'artagnan fought with Mr. de Jussac? A step there, a step here, then they ran around the tree and only then started stabbing each other. We won’t pull like that, but it also makes no sense to stab right away - it will not be clear.

So! In a combat system, a scoring ball will be calculated based on two groups of indicators:

Based on the listed indicators, a model will be built: the vertices of the graph will be all the organizations with which the borrower interacted in one way or another, the edges of the graph will have weight. The weight of the connection will be set in the range from 1 to 5, characterizing the degree of influence of the nodes on each other.

For example:

The degree of interaction and the interactions themselves will be determined, inter alia, using graph search algorithms.

In our test system, we will use the same topic with the musketeers and their connections. The model will be as close as possible to the combat and sufficiently demonstrate our idea. What will we ultimately come to, what will the model look like? Take your time to say: “Canalia!” Or “I do not need academies. Any Gascon from childhood is an academician! ” Everything will not be as primitive as it seems.

Short description: our musketeers decided to create a non-public joint-stock company (NPAO), which will supply jewelry and provide security services, they need a loan to start the activity. The credit institution is PJSC Korol, which commissioned the evaluation of NPO One for All

Features of the presented graph:

We have a lot of work ahead. Well, as part of this article, we are done. The stated objectives of the article, as it seems to us, have been achieved. We hope we managed to interest you, and you read to the end.


All Articles