Many attribute-based anonymous credential (ABC) schemes have been proposed allowing a user to prove the possession of some attributes, anonymously. They became more and more practical with, for the most recent papers, a constant-size credential to show a subset of attributes issued by a unique credential issuer. However, proving possession of attributes coming from K different credential issuers usually requires K independent credentials to be shown. Only attribute-based credential schemes from aggregatable signatures can overcome this issue. In this paper, we propose new ABC schemes from aggregatable signatures with randomizable tags. We consider malicious credential issuers, with adaptive corruptions and collusions with malicious users. Whereas our constructions only support selective disclosures of attributes, to remain compact, our approach significantly improves the complexity in both time and memory of the showing of multiple attributes: for the first time, the cost for the prover is (almost) independent of the number of attributes and the number of credential issuers. Whereas anonymous credentials require privacy of the user, we propose the first schemes allowing traceability. We formally define an aggregatable signature scheme with (traceable) randomizable tags, which is of independent interest. We build concrete schemes from the recent linearly homomorphic signature scheme of PKC 20. As all the recent ABC schemes, our construction relies on signatures which unforgeability is proven in the bilinear generic group model.