The necessity of determining the probability of default often conflicts with the requirement for employing parsimonious methodologies. The inclusion of a large number of regressors in Logit models or Machine Learning approaches can lead to overfitting, thereby introducing biases that distort the results. This study aims to examine the extent to which an indicator that synthesizes information related to balance sheet metrics can achieve a performance comparable to that obtained through a comprehensive set of indicators. In this paper, we introduce the Synthetic Performance Indicator (ISP), which is derived from specific balance sheet indicators. We demonstrate its effectiveness as a synthetic measure of financial stability in localized settings. Furthermore, we assess its potential to serve as a viable alternative to the broader panel of indicators from which it is constructed. Finally, we provide further evidence of how Machine Learning approaches, despite being effective in-sample, perform poorly out-of-sample.