Explanations　can　substantially　enhance　users’　trust　and　satisfaction　with　recommender　systems.　Counterfactual　explanations　have　demonstrated　remarkable　effectiveness　in　enhancing　the　performance　of　explainable　sequential　recommendation.　However,　existing　counterfactual　explanation　models　for　sequential　recommendation　ignore　temporal　dependencies　in　a　user’s　historical　behavior　sequence.　Moreover,　counterfactual　histories　must　be　as　close　as　possible　to　the　real　history;　otherwise,　they　will　violate　the　user’s　real　behavioral　preferences.　In　this　paper,　we　propose　Counterfactual　Explanations　with　Temporal　Dependencies　(CETD),　a　counterfactual　explanation　model　based　on　a　Variational　Autoencoder　(VAE)　for　sequential　recommendation　that　handles　temporal　dependencies.　When　generating　counterfactual　histories,　CETD　uses　a　Recurrent　Neural　Network　(RNN)　to　capture　both　long-term　preferences　and　short-term　behavior　in　the　user’s　real　behavioral　history,　which　can　enhance　explainability.　Meanwhile,　CETD　fits　the　distribution　of　reconstructed　data　in　a　latent　space,　and　then　uses　the　variance　obtained　from　learning　to　make　counterfactual　sequences　closer　to　the　original　sequence,　which　will　reduce　the　proximity　of　counterfactual　histories.　Extensive　experiments　on　two　real-world　datasets　show　that　the　proposed　CETD　consistently　outperforms　state-of-the-art　methods.　©　2023,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.