We got a message from a contestant who couldn't figure out why his algorithm didn't pass the beta cutoff. When we dug into it, we found a bug in the way we were calculating the cutoff beta value for the contest algorithms. It was one of those bugs that every programmer will recognize: we were using the last 13 trailing months to create the average, not the 12 months that we intended. The actual beta calculation was correct; it was the size of the window that was wrong.
We fixed the bug, and it changed the leaderboard. With this fix, 62 entries are no longer passing the 0.3 beta cutoff and 22 entries are now passing that weren't before. No entries in the top 30, and three entries in the top 60, will lose their beta badge. Two entries will move up into the top 25.
We apologize for the error and any unpleasant surprises. We appreciate your patience as we continually audit and improve the various moving pieces in the Quantopian Open. We very much appreciate the feedback we get that helps us find the bugs we didn't find ourselves.
You can check your entry’s rolling beta score with the following steps:
- Run a full minute-level backtest from 5/1/2013 to 4/30/2015.
- Click into the ‘beta’ tab on the left side of your full backtest
screen. - Take the mean of the twelve most recent values for twelve month
rolling beta (the rightmost column)