A Nonlinear Approach to assess the Risk Reward Ratio using the Machine Learning technique

In the recent years, the computational and statistic applications, especially regarding the machine learning technique, have seen an important and growing role in the academic debate. Our focus is toward Italian firms. In particular, the objective of our work is to assess the risk scores estimating the default probabilities. From the beginning of 1990s the machine learning techniques such as artificial neural networks have been successfully put into practice to bankruptcy prediction. One of the major critiques arisen from the risk assessment models applied to finance regards the assumption of normality, equally of covariance matrices and other modelling restrictions. In our work, we followed an alternative nonlinear separation method known as the Support Vector Machine (SVM) for the default risk analysis. We did not need any parameter restrictions nor prior assumptions in our estimates. Indeed, the current literature shows a strong evidence that the SVM approach outperforms the standard pattern for classification and regression function. Using the AIDA dataset of the Italian firms, we propose a SVM model based on the performance measures such as ratios of leverage, working capital, liquidity and activity in order to estimate the risk of bankruptcy. Finally, we compare the risk scores obtained for classification with the expected returns predicted by the SVM model. The results of our work point out that using nonlinear techniques for predicting bankruptcy allows to achieve better performances than traditional statistical ones and, moreover, shows the important predictors to estimate default probabilities.