solving convex MINLP problem


I am using second-order cone relaxation to convert my original non-linear problem. When I solve it using juniper (cplex as mip+ ipopt ), the solved objective is correct but takes longer than solving the non-linear problem itself for many test cases. However, I realized that for my convex MINLP problem I should use pavito (cplex as mip and ipopt as continuous) . But, my objective value is not even as good as the linearized problem. I am not sure what to make out of this. Do you think the formulation itself has an issue? Was my former way correct- using juniper to solve convex MINLP? I have very limited knowledge of the solvers.