Estimating the common agricultural policy milestones and targets by neural networks

The New Delivery Model, introduced by the 2023–2027 Common Agricultural Policy, shifts the focus of policy programming and design from a compliance-based approach to one based on performance. The objectives indicated in the national strategic plans are monitored through the definition of a set of milestones and targets. This makes it necessary to define realistic and financially consistent target values. The aim of this paper is to outline a methodology to quantify robust target values for result indicators. As the main method, a machine learning model based on multilayer feedforward neural networks is put forward. This method is chosen for its ability to model possible non-linearities in the monitoring data and estimate multiple outputs. The proposed methodology is applied to the Italian case, more specifically to estimate target values for the result indicator related to enhancing performance through knowledge and innovation for 21 regional managing authorities. The related performance is then compared with that of traditional methods adopted to estimate target values. Results demonstrate the superiority of neural networks and suggest that this methodology might be used as a tool to help all Member States fulfill the key task of setting coherent and realistic targets for all result indicators. •The New Delivery Model shifts the focus of the Common Agricultural Policy to a performance-based approach.•Multilayer feedforward neural networks are used to estimate multi-annual target values for result indicators.•Techniques of data augmentation and hyperparameter optimization of neural networks are adopted.•Neural networks ensure high prediction performance and high capability of generalization•The proposed methodology allows the definition of robust (i.e., realistic, consistent, and reliable) target values for result indicators.