A Latent Markov Approach for Clustering Contracting Authorities over Time Using Public Procurement Red Flags

Public procurement is a field particularly prone to corrupt acts. In this regard, red flag indicators are developed with the purpose of signalling anomalies and alerting the system for a potential risk of corruption. By exploiting data coming from the Italian National Database of Public Contracts, a set of red flag indicators for the measurement of the risk of corruption in public procurement are computed for a sample of contracting authorities and monitored over time. Afterwards, a latent Markov model for continuous responses is applied to the data at issue, in order to: i. identify groups of contracting administrations (i.e., the model latent states), characterised by different behaviours on the basis of the selected red flags, and ii. estimate the transitions over time across the ascertained groups. Results show that seven profiles of administrations may be highlighted on account of the selected red flags. Among them, four profiles include contracting authorities with indicator values associable to a low risk, whereas one profile can be labelled as the most at-risk one according to the respective indicator means.