Flexible estimation of time-varying effects for frequently purchased retail goods: a modeling approach based on household panel data