The Use of a Genetic Algorithm in Forecasting Air Carrier Financial Stress and Insolvency

While statistical and artificial intelligence methods such as Artificial Neural Networks (ANN) have been used successfully to classify organizations in terms of solvency or insolvency, they are limited in degree of generalization either by requiring linearly separable variables, lack of knowledge of how a conclusion is reached, or lack of a consistent approach for dealing with local optimal solution whether maximum or minimum. This research explores the use of a method that has the ability of the ANN method to deal with linearly inseparable variables and incomplete, noisy data; and resolves the problem of falling into a local optimum in searching the problems space. The paper applies a genetic algorithm to a sample of U.S. airlines and utilizes financial data from carrier income statements and balance sheets and ratios calculated from this data to assess air carrier solvency. Introduction The forecasting of financial stress and bankruptcy in the U. S. airline industry has become quite important over the past decade, as the number of carrier insolvencies has increased steadily in the years after the deregulation of the industry. Previous research studies by the authors [Chow, Gritta, 1991; Davalos, Gritta, 2002; Goodfriend, Gritta, 2004; Gritta, Davalos, 2003} over the years have utilized different techniques such as Logit, MDA (Multiple Discriminant Analysis), and ANNs (Artificial Neural Networks). And while these statistical and artificial intelligence methods have been used successfully to classify airlines in terms of solvency or insolvency, they are limited in degree of generalization either by requiring linearly separable variables, lack of knowledge of how a conclusion is reached, or lack of a consistent approach for dealing with local optimal solution whether maximum or minimum. This research study explores the use of a method that has the ability of the ANN method to deal with linearly inseparable variables and incomplete, noisy data, and does resolve the problem of falling into a local optimum in searching the problems space.