Unless... is the paper trading correlation the daily correlation only during the contest? (ie: it just went from 8 data points to 9) I had assumed you were using higher-frequency data, but that would explain it.

Drawdown seems to use daily data, so the same is probably true of correlation.

Also, when the contest calculations are run more than once (after market close in the evening due to new April backtests coming onboard), the next run also affects March contest scores, resulting in changes to those scores and standings.

LOL really? As though perhaps the same day is getting counted twice?

Simon - yes, indeed, all of the Leaderboard calculations are based on end-of-day results. The early days of the contest are often very noisy, particularly around drawdown and Calmar ratio because of that. They smooth out some after a full month. By the end of March the swings will be more muted day-to-day.

(Obviously, there are limitations to how smooth they can possibly be inherent in the choice of a one-month minimum contest, but we're accepting those limitations as a trade-off to the benefits to a fast-moving contest).

Gary et al - The March and April contests are calculated entirely independent of each other. Obviously, entries in the March contest are duplicated in the April contest. Because we're using a ranking mechanism based on the pool of entries, a given set of metrics in March might rank either higher or lower in April, depending on the strength of the rest of the pool in the given contest.

In general, we run the leaderboard calculations once per day and then publish the results. Sometimes we get "new data" later than night and we re-run the leaderboard calculation. "New data" for the contest with open submissions is often newly-completed backtests or something of that nature. Other times, our process might initially omit a submission because of some error, but when we dig into it, we find the error is in our calculation and then we put the submission back into the pool. Most of the time that's an invisible process, but because this is a rank-based system, everyone is affected at least a tiny bit when the pool gets updated. That's why you might see more than one update to the leaderboard in a night.

We're definitely still learning and refining our process. One philosophy we could use for this is to never update the leaderboard until we're sure it's right. Instead, we've chosen to update the leaderboard when we have a solid-but-imperfect update, and then correct the errors later that night. However, on the day that really matters, the last day, we have a more rigorous verification process before we update the leaderboard.

Dan, I have an algo running for the contest and due its parameters it did not trade yet, a 100% cash position. I'm not appearing in the leaderboard though. 100% cash is a position too as my algo waits till it gets things on a discount, so I would expect to be at least mentioned on the leaderboard....

Peter, it looks like you entered on March 10th, which puts you in the April contest but not the March contest. (March deadline was March 2, 9:30AM). The entry I'm looking at is on the April board here: https://www.quantopian.com/leaderboard/54fee4e412b2145de00005d8?month=4&year=2015

If there is a different entry that's missing, please drop us a note at [email protected] with some details so we can figure it out.

As a side note - we agree that a patient algorithm is a real strategy, and can be a very good strategy. A patient algorithm might not win at first. That's why we're doing the rolling entry the way that we are. The tortoise might beat the hare as we look at longer time horizons.