Peer to Peer systems have shown to be very powerful to build very large scale distributed information systems. They are self organized, and provide very high availability of the data. However, the management of malicious peers is a very open problem for the Peer to Peer research community, and building trust is a very difficult task. In this context, Reputation Systems have shown to be a very good solution to build trust in Peer to Peer systems. Nevertheless, using only the reputation value of a peer to decide to make a transaction is not sufficient to guarantee that it will succeed, and the use of the credibility of recommendation emitters does not always significantly mitigate the computed reputation. We show in this paper the importance of the notion of risk associated to the reputation value, and why a better decision can be taken using both, the reputation and a risk value, for a given peer. We present some metrics based on the list of recommendations for a peer that allow to detect some suspicious behaviours that can alert the application of the presence of a malicious peer. The proposed metric is flexible such that an application can adapt the metric to its needs, given more or less weight to some specific types of behaviours. We present some simulations to show the influence of malicious behaviours of a peer over its reputation value with the evaluation of the associated risk, and how our metric can detect this kind of behaviours. We conclude about the need to use a risk factor associated to the reputation value, and present some future works about the risk metrics.