Which Characteristics Predict the Survival of Insolvent Firms? An SME Reorganization Prediction Model

Abstract

The negative impact of insolvency, especially in small and medium enterprises, informs the objective of this paper: to study the characteristics of bankrupt firms to achieve a preventive diagnosis for reorganization by means of artificial intelligence () methodologies such as rough set and methods. The models obtained show not only the key variables to predict insolvency, but also their relations and the critical values. Using only five firm characteristics (sector, size, number of shareholdings, return on assets, and cash ratio), our model could reduce delays and costs, since it is able to predict which firms will undergo reorganization or liquidation before the legal procedure.

Notes

1. In fact, the 23 million SMEs in the EU represent 99 percent of businesses, and are a key driver for economic growth, innovation, employment, and social integration. http://ec.europa.eu/enterprise/policies/sme/index_en.htm

2. The terms “insolvency,” “business failure,” and “financial distress” are used in the same way: when the firm has no cash to pay its debts.

3. For a review of all these methodologies see Kumar and Ravi (2007).

4. SABI (Iberian Balance Sheet Analysis System) is a product of Bureau Van Dijk. For more information: http://www.informa.es/en/financial‐solutions/sabi.

5. To classify the companies’ sector the National Statistics Institute codification has been followed (CNAE‐2009, http://www.ine.es/jaxi/menu.do?L=1&type=pcaxis&path=%2Ft40%2Fclasrev%2F&file=inebase).

6. Rough set analysis was performed using RSES2, developed by the Institute of Mathematics, Warsaw, Poland. http://logic.mimuw.edu.pl/~rses/

7. http://www.cs.waikato.ac.nz/ml/weka/

8. Cross‐validation comprises several training and testing runs. The data set is first split into several, possibly equal in size, disjointed parts. Then, one of the parts is taken as a training sample and the remainder (sum of all other parts) becomes the test sample. The classifier is constructed by means of the training sample and its performance is checked against the test sample. These steps are repeated as many times as there are data parts, so that each of the parts is used as training set once. The final result of the cross‐validation procedure is the average of scores from subsequent steps.

Additional information

Notes on contributors

Maria‐del‐mar Camacho‐miñano

Maria‐del‐Mar Camacho‐Miñano is assistant professor at Department of Financial Economics and Accounting II, Complutense University of Madrid.

Maria‐jesus Segovia‐vargas

Maria‐Jesus Segovia‐Vargas is assistant professor at Department of Financial Economics and Accounting I, Complutense University of Madrid.

David Pascual‐ezama

David Pascual‐Ezama is assistant professor at Department of Financial Economics and Accounting II, Complutense University of Madrid.