The problem of variable selection for financial distress: applying GRASP methaeuristics

We use the GRASP procedure to select a subset of financial ratios that are then used to estimate a model of logistic regression to anticipate financial distress on a sample of Spanish firms. The algorithm we suggest is designed "ad-hoc" for this type of variables. Reducing dimensionality has several advantages such as reducing the cost of data acquisition, better understanding of the final classification model, and increasing the efficiency and the efficacy. The application of the GRASP procedure to preselect a reduced subset of financial ratios generated better results than those obtained directly by applying a model of logistic regression to the set of the 141 original financial ratios.