We develop a trading strategy random selection approach that allows us to generate many trading strategies: some of which have already been studied and published, some have likely been studied but not published, and some that have never (and likely never will) be studied. Using the large cross-section of trading strategies we construct precise multiple testing hypothesis adjusted p-values (and t-statistics) that account for the cross-correlation in the data used to generate such strategies (i.e., accounting variables and returns data). After taking such adjustments into account, most strategies that have already been  studied would not be statistically significant (even in-sample), and the ones that would be significant, and that have yet to be studied, could be dismissed based on economic considerations. Joint with with Tarun Chordia and Alessio Saretto.