A neural network model to predict business failure in construction companies, in the United States of America

The construction industry has one of the highest rates of bankruptcy in the United States of America. Although there are many generic prediction models developed to help company managers to predict whether their companies are still healthy or will fail, there was not a specific model trained and tested just using data from construction companies (heavy, utility and commercial construction). The purpose of this dissertation was to create a model using neural networks that was able to predict business failure in construction companies one, two and up to three years before it happened. Data from sixty-seven healthy and bankrupt companies were collected. Although twenty-six financial ratios were first calculated, seven ratios were found to be the most significant indicators and were used to train and test the neural networks. Three neural networks (one, two and three years prior to business failure) were trained and tested. In order to understand the importance of the results, data from randomly chosen construction companies were entered into Altman's model, which is a generic predictor of business health. The results obtained using the neural network models were more accurate than those obtained using Altman's model. Afterwards, a numerical analysis was performed to identify which of the financial ratios were the most important. The results showed that the debt-to-equity ratio, debt-to-assets ratio and the gross profit margin ratio could generate higher changes to the financial condition of a construction company. It was hoped that the results obtained in this dissertation showed that future development of this models could become an important tool for construction companies.