CFE 2019: Start Registration
View Submission - CMStatistics
Title: Spatial dependence and localization in bankruptcy prediction: A comparative analysis on Italian manufacturing firms Authors:  Maria Simona Andreano - Universitas Mercatorum (Italy) [presenting]
Roberto Benedetti - University of Chieti - Pescara (Italy)
Federica Piersimoni - ISTAT (Italy)
Andrea Mazzitelli - Universitas Mercatorum (Italy)
Abstract: The interest in prediction of firms bankruptcy is increased in recent years, when recession has sharply raised the number of distressed manufacturing firms. Although numerous studies focus on this topic and several attempts to provide a solution for the problem, predicting financial firms stress is not a trivial task. The most popular parametric models applied by bankruptcy researchers are the Logit and Probit, where failure is seen as a dichotomic event, whereas recently, failure prediction methods moved to more comprehensive machine learning techniques. The space and the location of firms have been considered a decisive factor in many fields of business-related research, however they are rarely applied in bankruptcy analysis. A spatial econometric methodology is applied to evaluate the effect of geographical location on the probability of business failure. Moreover, spatial dependence is included in machine learning techniques through an Iterated Conditional Modes (ICM) algorithm, originally introduced on image processing, based on non-degenerate Markov random fields. Non-spatial and spatial models are applied on about 12.000 manufacturing firms, located in Central Italy. Different evaluation metrics are selected to compare the performance of the various approaches. Our application shows that spatial contagion effects are an important issue when modelling bankruptcy probability and spatial models outperform classical ones.