Search

Company managers, financial institutions and insurers are constantly concerned about the solvency of their corporate clients. The increased demand for solvency analysis and higher computing capacity are driving the adoption of non-traditional methods. The present study examines firm insolvency from a geographically concentrated perspective, focusing on small and medium-sized enterprises in the Pécs and Budapest city regions. It builds on artificial intelligence and machine learning methods and provides a comparative analysis of several aspects of neural network, SVM, bagging, and random forest meta-methods. The authors finally attempt to open the 'black-box' phenomenon by assessing the importance of explanatory variables and identifying the most significant financial explanatory variables. [ABSTRACT FROM AUTHOR]

Due to technological advances, personal computers have made the application of machine learning methods characterised by high computational requirements, possible. In line with this, significant progress has been made in the functionality of open source statistical and data analysis software. In spite of these favourable processes, the number of Hungarian studies using machine learning methods lags behind that of articles published in the international literature. One possible reason for this may be that machine learning-based modelling requires a high level of programming skills. The aim of this article is to overcome this difficulty by presenting such an R package that allows model development using machine learning, without prior programming knowledge. [ABSTRACT FROM AUTHOR]

The Blinder-Oaxaca decomposition splits the average gender wage gap into two parts: one that is explained by covariates (composition effect) and another that is unexplained (wage structure effect). In this study, male and female wage functions are decomposed by OLS and random forest regressions, and the results obtained on a Hungarian dataset are compared. It is found that the composition effect is larger in random forest regression than in OLS regression, resulting in smaller wage structure effects; however, these latter estimated by the two methods are almost identical. Individual wage structure effects have low correlation, i.e. there can be significant differences among them. Descriptive statistics of covariates show differences between the two methods, which cannot be traced back to one single covariate. The author's conclusions confirm the results of Weichselbaumer-Winter-Ebmer [2005] who argued that different methodologies of the Blinder-Oaxaca decomposition estimate the same composition and wage structure effects on average but the methodology significantly influences the importance of each variance. [ABSTRACT FROM AUTHOR]