SciCast’s unique feature is the ability to make conditional edits, also called assumptions or combinatorial edits. Now you’ll have a chance to compete for big prizes for your making at least at least 25% of your forecasts conditional forecasts. combo edits!
We’ll be offering $16,000 in prizes for conditional forecasts only made from April 23 to May 22. Adding a link requires making a conditional forecast, and therefore counts as a conditional forecast.
Check out eligible questions with this symbol All questions are eligible. Note that The “Combo Contest” Leaderboard will be available starting around May 1, and will be used for payout.
.
Users who have LIKED this post:
Pingback: Overcoming Bias : SciCast Contest
It is stupid to have the “When will flu season end in region X” questions in this contest. These have no existing links from other questions that have yet to resolve. The only plausible conditioning questions are other similar questions. But linking between two multinomial questions with so many outcomes requires essentially infinite point commitment. So you will not see any new links made, and thus no contest-eligible conditional edits can be made. I suggest replacing these questions with other, randomly selected ones.
@sflicht: (1) Linking two of them is actually pretty cheap, but they are already all paired. A second link is expensive but do-able. The third link is unaffordable. (2) But you don’t need to add links to earn points, just make combo edits. (3) These questions had the huge advantage that they would resolve by end of contest. (4) But for reasons pointed out by @Faber, we had to change the contest rules. To avoid ratchets, we had to include all questions. So, your wish is mostly granted.
I think the algorithm for pricing the links should not depend upon existing links to questions that are no longer editable — there isn’t (or at least shouldn’t be) any ongoing computational cost associated to maintaining those links.
Link Price does not depend on resolved questions. It does consider paused questions, because they could be unpaused at any time. Are you seeing something that looks different?
“Paused” questions that are actually past the last-trade-date cannot be unpaused. E.g. the “when will flu season peak” questions link to the “when will flue season end” questions. The former are all waiting to be judged — they will not be unpaused, just final-resolved.
They are in the Bayes net until final resolved, and therefore affect inference time. Although you’re right that an admin isn’t going to unpause these, technically they could do so. Final resolution is what clears them out of the engine. (The core problem with these questions is the number of states per question. I find them unwieldy, but the authors needed both week-level resolution and a distribution on the values, and we don’t have a production version of a hierarchically-structured ordered node.)
Pingback: SciCast Contest | I World New
Pingback: Updates to the Combo Edits Contest | The Official SciCast Blog
1. You can dump out-of-the-money points into real-money points by trading against yourself. (Glad this was addressed)
2. You can make all your trades conditional by just conditioning on some 100% likely contract, thereby defeating the mandate of conditional trades.
Yes, you can make frivolous conditional edits. That one we recognized from the beginning. But our secret goal is to see what kind of mutual information develops if we increase the conditional forecasts. I expect to get some frivolous conditionals, because people will want to ensure they’ve passed threshold. But if we get only frivolous conditionals, then it we have to consider whether there are no real relationships among our questions, or expressing them is still too difficult / obscure, or that maybe in practice conditionals aren’t worth their weight.
The blog post indicates that 25% of forecasts must be “conditional” in order to qualify for the contest, while the “Official Rules” indicates 50%. Which of these is incorrect?
25% is correct — I thought we had fixed that everywhere. Thanks for pointing it out.
Did the contest end (at a random time) on the 21st or the 22nd?
The rules specify the 21st, but I just noticed that the above text says the 22nd.
The 22nd. Sorry for the inconsistencies.
Pingback: Finding Answers in the Bitcoin Block Size Debate Using Prediction Markets | CryptoCurrency Buzz
Pingback: Announcing the Combo Edits Contest Winners! | The Official SciCast Blog