Explainable Recommendation Systems (RS) enhance the user experience on online platforms by recommending personalized content, as well as explanations for the given recommendations to add transparency and build up trust in the platforms. Extending the notion of explainable RS, in this paper we define Why-Not explanations for recommendations that were expected but not returned, and propose and implement a technique for computing Why-Not explanations in a post-hoc manner for a graph-based RS. Our approach builds on the notion of counterfactual explanations in the means of a set of user-rooted edges to add or remove in the graph, in order to place the missing recommendation to the top of the recommendation list, and provides in this way actionable insights on the source data and their interrelations. Our experimental evaluation on a real-world data set demonstrates the feasibility of our proposal and reveals interesting directions for future work.