Optimal trade-off between sample size and precision of supervision for the fixed effects panel data model