A battle of pooled and panel data in credit risk modelling