ABSTRACT
From a methodological consideration, focusing on performance opposed to distress, the concern is primarily with prediction accuracy and stability when using future data. Hence, the robustness of the model is based on its ability to detect subtle variations in airlines’ operating characteristics that can cause shift from loss to profit or vice versa. To capture these variations the research presented in this chapter compares two approaches, namely a single year and a multi-year training set: a single-year training set is the tradition in distress prediction. Since a performance prediction model is not focused on bankruptcy (termination event) or distress (not meeting obligations) the performance prediction model has to capture the dynamism of airline operations from year to year and its influence on performance. A failure prediction model based on financial ratios does not indicate “what” has gone wrong if a company fails, because financial ratios are symptomatic rather than causal by nature. Examples exist of managers trying to improve “the ratios” without addressing underlying problems (Altman, 1993). Hence, it is of interest to develop prediction models based on variables that reflect better operational decisions. Using non-financial variables can accomplish this to a degree. Financial distress prediction models are generally considered of little use in multi-country samples due to differences in accounting laws, and differences in the business and political climate of different countries (ibid.). Again, national biasing factors can be partially controlled by focusing the models on nonfinancial variables. What is more, in the case of airlines often there is only one airline or a few airlines operating in a single country making an industry-specific prediction model for one country impossible. Using multi-year data to construct prediction models is still relatively underexplored. Berg (2006) developed a multi-year model, and compared it with a one-year model and found that the multi-year model exhibited robustness to yearly fluctuations that was not present in the one-year model. Similar results were reported by Vieira et al. (2009) using three years of data applying several different methodologies including a multi-layered neural network. Consequently, a multi-year data model was expected to capture better the year-to-year changes causing fluctuations in airline profitability. Therefore, the objective in this research was to construct a neural network performance prediction model, a multi-country single industry model and to compare a multi-year and single year model. The industry is international airlines, the data used is non-financial, and input parameters were selected on significant difference of the means between the two performance states. In the following sections an overview is given of performance measurement, distress prediction and airline industry-specific prediction models. Then the methodology is explained, followed by research findings, and conclusions.
