Abstract
This paper contributes to the analysis of quantitative indicators (i.e., red flags or screens) to detect corruption in public procurement. It presents an approach to evaluate corruption risk in public tenders through standardized ML tools applied to detailed data on the content of calls for tenders. The method is applied to roadwork contracts in Italy and three main contributions are reported. First, the study expands the set of commonly discussed indicators in the literature to new ones derived from operative practices of police forces and the judiciary. Second, using novel and unique data on firm-level corruption risk, this study validates the effectiveness of the indicators. Third, it quantifies the increased corruption-prediction ability when indicators that are known to be unavailable to the corruption-monitoring authority are included in the prediction exercise. Regarding the specific red flags, we find a systematic association between high corruption risk and the use of multi-parameter awarding criteria. Furthermore, predictability of the red flag makes them ineffective as prediction tools: the most obvious and scrutinized red flags are either uncorrelated with corruption or, even, negatively associated with it, as it is the case for invoking special procedures due to "urgency," or the extent of publicity of the call for tender.