Bayesova logistička regresija za predviđanje stečaja trgovačkih društava u djelatnosti trgovine na veliko i malo

master's thesis

• Title (english): Bayesian logistic regression for predicting bankruptcy of wholesale and retail trading companies
• Author: Matteo Moćan
• Mentor: Josip Arnerić (mentor)
• Committee member: Andrija Sabol (član povjerenstva)
• Committee member: Ena Pecina (član povjerenstva)
• Granter: University of Zagreb; Faculty of Economics and Business; (Department of Managerial Economics); Zagreb
• Defense date and country: 2023-09-14, Croatia
• Scientific / art field, discipline and subdiscipline: SOCIAL SCIENCES; Economics; Economics of Entrepreneurship

Abstract

Pojavom COVID-19 krize, naglo se poljuljala svjetska ekonomija, te se isto dogodilo i Hrvatskoj. Takve okolnosti dovele su u pitanje preživljavanje poduzeća, naročito trgovine koja je pretrpjela najviše uvođenjem protuepidemijskih mjera. Stoga je od velike koristi s velikom preciznošću predvidjeti stečaj trgovina u navedenim uvjetima, kako bi se ispravnim odlukama spriječio stečaj i kako bi se u idealnoj situaciji stekla konkurentska prednost. Zato se provodi stratificirano uzorkovanje poduzeća kako bi omjer uspješnih i neuspješnih trgovačkih društava u sektoru trgovine na veliko i malo odgovarao populaciji s obzirom na njihovu heterogenost, te kako bi se pritom odredio optimalni prag vjerojatnosti za potrebe klasifikacije. Oblikuje se i razvija model standardne i Bayesove logističke regresije s jednim financijskim pokazateljem iz svake grupe financijskih pokazatelja koji su se pokazali signifikantnima za predviđanje stečaja trgovačkih društava djelatnosti trgovine na veliko i malo; koeficijent tekuće likvidnosti, koeficijent zaduženosti, koeficijent obrtaja kratkotrajne imovine i EBIT marža. Cilj je utvrditi koji financijski pokazatelji najbolje diskriminiraju poduzeća na uspješna ili neuspješna i pridonosi li većoj prediktivnoj sposobnosti Bayesov pristup logističkoj regresiji u uvjetima neizvjesnosti. Koeficijent obrtaja kratkotrajne imovine najbolje diskriminira između aktivnih poduzeća i onih u stečaju, dok EBIT marža najmanje. Nadalje model logističke regresije predviđa s 90,01% točnosti, a model Bayesove logističke regresije s 90,22% točnosti. Takve performanse modela ukazuju na veću preciznost Bayesove logističke regresije, te da je ona prikladnija za predviđanje stečaja u uvjetima neizvjesnosti. Model Bayesove logistička regresije je dvije godine prije stečaja trgovačkih društava predvidio stečaj sa 87,47% točnosti i 76,45% točnosti tri godine prije stečaja.

Abstract (english)

With the emergence of the COVID-19 crisis, the world economy suddenly shook, and the same thing happened in Croatia. Such circumstances brought into question the survival of companies, especially retail, which suffered the most due to the introduction of anti-epidemic measures. Therefore, it is very useful to predict the retail bankruptcy in the mentioned conditions with great precision, in order to prevent bankruptcy with the correct bankruptcy decisions and to gain a competitive advantage in the ideal situation. That is why a stratified sampling of companies is carried out so that the ratio of successful and unsuccessful companies in the wholesale and retail trade sector corresponds to the population with regard to their heterogeneity, and in order to determine the optimal probability threshold for the purposes of classification. A standard and Bayesian logistic regression model is designed and developed with one financial indicator from each group of financial indicators that have proven to be significant for predicting the bankruptcy of wholesale and retail trade companies; current liquidity ratio, debt ratio, short-term asset turnover ratio and EBIT margin. The goal is to determine which financial indicators best discriminate between successful and unsuccessful companies and contribute to a greater predictive ability of the Bayesian approach to logistic regression under conditions of uncertainty. The current asset turnover ratio discriminates best between active and bankrupt companies, while EBIT margins do the least. Furthermore, the logistic regression model predicts with 90.01% accuracy, and the Bayesian logistic regression model with 90.22% accuracy. Such performance of the model indicates greater precision of Bayesian logistic regression, and that it is more suitable for predicting bankruptcy under conditions of uncertainty. The Bayesian logistic regression model predicted bankruptcy two years before bankruptcy of companies with 87.47% accuracy and 76.45% accuracy three years before bankruptcy.

• Keywords: standardna logistička regresija; Bayesova logistička regresija; predviđanje stečaja; financijski pokazatelji; trgovina na veliko i malo
• Keywords (english): standard logistic regression; Bayesian logistic regression; bankruptcy prediction; financial indicators; wholesale and retail trade
• Language: croatian
• URN:NBN: urn:nbn:hr:148:259183
• Study programme: Title: Business Economics; Study programme type: university; Study level: integrated undergraduate and graduate; Academic / professional title: magistar/magistra ekonomije (magistar/magistra ekonomije)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2023-09-18 06:57:21