We propose a formal definition of backtestable statistic: a backtest is a null expected value involving only the statistic and its random variable, strictly monotonic in the former. We discuss the relationship with elicitability and identifiability which turn out being necessary conditions. The...
