Algoritmos de Random Forest como alerta temprana para la predicción de insolvencias en empresas constructoras = Random Forest algorithms as early warning tools for the prediction of insolvencies in construction companies

Abstract

The European Union's concern with preventing companies from reaching insolvency proceedings motivated the enactment of Directive (EU) 2019/1023 of the European Parliament and of the Council, and its mandatory transposition into Member States' regulations by July 17, 2021. This Directive states that debtors must have access to early warning tools to detect situations of J. I. Sordo Sierpe, M. del Río Merino, Á. Pérez Raposo, V. Vitiello 10 Anales de Edificación, Vol. 7, No 1, 9-18 (2021). ISSN: 2444-1309 imminent insolvency. This research aims to contribute to the development of such early warning tools for a very specific sector: residential and non-residential construction. The methodology has been divided into two phases, each with its own specific objective: (1) to select the predictor variables that can best explain the model (traditional statistical techniques have been used for this purpose); and (2) to select the algorithms that provide the greatest precision for the early warning tool model from among five Random Forest algorithms. The main objective of this is to obtain warning signs sufficiently enough in advance that insolvency situations can be detected. The fundamental aim is to achieve a model without using the profit and loss accounts from the construction companies under investigation. This is so to avoid the lack of objectivity that income, and therefore accounting results, may have in this sector. Accuracy percentages of over 85% were obtained three years before insolvency occurred using only balance sheet ratios. The main value is to be able to apply the early warning tool in a simple way, using little amounts of data, especially for the debtor, who can react early enough to avoid a potentially irreversible financial situation.