Analiza mjere pogreške predviđanja rizika modelom binomne logističke regresije

master's thesis

• Author: Toni Ruščić
• Mentor: Vesna Lužar-Stiffler (mentor)
• Committee member: Vesna Lužar-Stiffler (predsjednik povjerenstva)
• Committee member: Bojan Basrak (član povjerenstva)
• Committee member: Vanja Wagner (član povjerenstva)
• Committee member: Vjekoslav Kovač (član povjerenstva)
• Granter: University of Zagreb; Faculty of Science; (Department of Mathematics); Zagreb
• Defense date and country: 2019-07-16, Croatia
• Scientific / art field, discipline and subdiscipline: NATURAL SCIENCES; Mathematics

Abstract

U ovom diplomskom radu cilj je bio analizirati mjere pogreške predikcije rizika u bankama i kreditnim institucijama. Za dobivanje rezultata koristili smo metodu binomne logističke regresije. U prvom dijelu diplomskog rada opisana je povijest kreditnog bodovanja, statističke i nestatističke metode koje se mogu koristiti za predikciju rizika te općenito o generaliziranim linearnim modelima i logističkoj regresiji. U drugom dijelu diplomskog rada opisan je Monte Carlo eksperiment pomoću kojeg smo ispitali mjere pogreške predviđanja rizika pri raznim uvjetima kao što su distribucija prediktorskih varijabli, broj prediktorskih varijabli, veličina uzorka i korelacija između varijabli. Rezultati su pokazali da su najbolji modeli Gamma distribucije s većim parametrom oblika k te da korelacija jako utječe na statističke modele. Modeli u kojem su varijable neovisne te modeli s više prediktorski varijabli su pokazali bolju prediktivnu sposobnost.

Abstract (english)

In this thesis, the aim was to analyze risk measures in banks and credit institutions. We used the binomial logistic regression method to obtain results. The first part of the thesis deals with the history of credit scoring, statistical and nonstatistical methods that can be used for risk prediction and about generalized linear models and logistic regression in general. In the second part of the thesis, we have described the Monte Carlo experiment by which we tested risk estimation measures at various perspectives such as probability distribution of predictor variables, number of predictor variables, sample size, and correlation between variables. The results showed that the best models are when predictor variables have Gamma distribution with the greater shape parameter k and that the correlation strongly influences the statistical models. Models in which variables are independent and models with more predictor variables have shown better predictive capability.

• Keywords: binomna logistička regresija; kreditno bodovanje; predikcija rizika
• Keywords (english): binomial logistic regression; credit scoring; risk prediction
• Language: croatian
• URN:NBN: urn:nbn:hr:217:099842
• Study programme: Title: Finance and Business Mathematics; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra matematike (magistar/magistra matematike)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2019-12-02 12:27:50