Probabilistic networks for bankruptcy risk prediction: a pilot study on Italian farms

The development of models for bankruptcy risk prediction has gained much attention in recent years due to the unprecedented availability of financial statement data. However, existing models lack of the flexibility required to adequately satisfy general user queries. For example, most of them have a fixed prediction horizon and/or cannot manage incomplete evidence. Probabilistic Networks (PNs) can overcome these limitations because they formalize the joint distribution of all variables instead of simply the conditional distribution of the response. However, just a few studies have adopted PNs so far and none of them has exploited their higher flexibility compared to conventional methods. Our objective is to illustrate the practical advantages of bankruptcy prediction models based on PNs, and to present the results of a pilot study on Italian farms showing that PNs can achieve a predictive power in line with the one of conventional methods. As such, we believe that the use of PNs can substantially improve the effectiveness of bankruptcy prediction models in supporting decisions.