Predictive Power Measures for Scoring Models

KOLÁČEK, Jan and Martin ŘEZÁČ. Predictive Power Measures for Scoring Models. In Výjezdní zasedání CJH Liberec. 2010.

Basic information

• Original name: Predictive Power Measures for Scoring Models
• Name in Czech: Prediktivní míra síly skóringových modelů
• Authors: KOLÁČEK, Jan and Martin ŘEZÁČ.
• Edition: Výjezdní zasedání CJH Liberec, 2010.

Other information

• Original language: English
• Type of outcome: Presentations at conferences
• Field of Study: 10103 Statistics and probability
• Country of publisher: Czech Republic
• Confidentiality degree: is not subject to a state or trade secret
• Organization unit: Faculty of Science
• Keywords (in Czech): credit scoring; indexy kvality; Giniho index; lift; integrální relativni lift; informační hodnota
• Keywords in English: Credit scoring; Quality indexes; Gini index; Lift; Integrated Relative Lift; Information value

Abstract

Credit scoring models are widely used to predict a probability of an event like client's default. To measure the quality of the scoring models it is possible to use quantitative indexes such as Gini index, K-S statistics, C-statistics and Lift. They are used for comparison of several developed models at the moment of development as well as for monitoring of quality of those models after deployment into real business. The paper deals with mentioned quality indexes, their properties and relationships. The main contribution of the paper is proposition and discussion of indexes and curves based on Lift. Curve of ideal Lift is defined, Lift ratio is proposed as analogy to Gini index. Integrated Relative Lift is defined and discussed. Also the problem of estimation of Information value is mentioned. Commonly it is computed by discretisation of data into bins using deciles. One constraint is required to be met in this case. Number of cases have to be nonzero for all bins. If this constraint is not fulfilled there are numerous practical procedures for preserving finite results. As an alternative method to empirical estimates we can use the kernel smoothing theory. Finally, a new approach to measure power of scoring models is discussed and a new quality index is proposed. A simulation study compares it with other quality indexes.

Abstract (in Czech)

K posouzení kvality skóringových modelů se používají kvantitativní indexy jako např. Giniho index, K-S statistika, C-statistika, Lift a informační hodnota. Práce se zabývá zmíněnými indexy, jejich vlastnostmi a vztahy mezi nimi. Hlavním přínosem této práce je navrhnutí nových indexů a křivek založených na Liftu. Je definována křivka ideálního Liftu a Lift ratio jako analogie k Ginimu koeficientu. Dále je také navržen a diskutován integrální relativní Lift a jsou popsány problémy při odhadování informační hodnoty. Nakonec je navržen nový přístup k problematice měření kvality skóringových modelů a také nový index kvality, který je porovnán s ostatními indexy v simulační studii.

Links

• LC06024, research and development project: Name: Centrum Jaroslava Hájka pro teoretickou a aplikovanou statistiku

Investor: Ministry of Education, Youth and Sports of the CR, Jaroslav Hájek Center for Theoretical and Applied Statistics