InfernalMachine's dashboard

I like it.

Here's another idea. When a question is created, the creator doesn't set the initial odds. Instead, the question is put (at first) into New Category. These questions allow people to select an option but not bet. They are prominently displayed on the home page, where people can look at them, vet them for ambiguity and crappiness and flag them as such if required, and also click on their personal guess/choice at the correct answer (but ONLY once per question). When 100 people have voted and less than 7 or whatever have flagged, it goes on to the next stage. That might be a vetting by an editor or a group, but maybe just an ed to check for flags and comments explaining the problem. If it passes this check (or had no flags), it goes live, into its specified category. If it fails an email is sent to the creator telling them to fix it and resubmit it. Now, where do the odds come from? The hundred people who voted on one of the options. (It's a different prediction market , where you can only vote $1 once, but it forms a rough approximation of the odds. The creator can't vote on their own question, just the first 100 people who happen upon it (no repeat voting on the same question, but encouraged to vote once on many new questions, in fact the $20 we get daily could be replaced with $1 per new question vetted and voted on.

Should solve the odds gaming problem - in fact deliver a better approximation than any individual could- as well as help with the voiding problem.

Two other gaming problems are cashing in after the event is determined and the weird oscillations when someone puts a huge bet into a thin market.

The first can be solved by either making cashing out end when betting is suspended, or by having the creator set a fixed duration after the suspension where cashing in is still allowed. But it can't stay open until the settlement without creating a sense of unfairness. (If any of my gains have been ill-gotten, it is in this area. By "unbetting" after betting is over you can make use of the differential between the actual worthlessness of your prediction and a sleepy market's failure to drive your bet's current value to $0 to save yourself a few squandered dollars. If the market keeps up with the facts though, cashing out at the bitter end is not cheating, it's pointless.)

The oscillation problem can be fixed by having a limit on a bet relative to the current capitalization of the market. It could be low (10-25%) or high (90%-200%) or somewhere in between. You might also want to have a set dead-time where you can't re-bet or cash in that bet (5 minutes, an hour, not much more than that) which allows the market to swallow the bet and restabilize itself.

This approach might make these 12 hour freezes unnecessary, as well as having certain other benefits.

What say the rest of you? Is my idea doable, easily incorporated into the site programming?

It's not so bad if questions are created 24+ hours in advance.

"Learn the game" sounds like an idea.

Not sure about "inordinate returns". In the last 30 hours I tripled my worth, and if a question hadn't been settled earliy and potentially incorrectly (dem super tuesday delegate winner), I'd have more than the the current leader on the board and have 5x my wealth from 30 hours ago, I didn't game anything, but I did obsess, did enormous research, glued myself to the computer and tv for many hours, and day-traded like the in-play questions for the super bowl. Crazy. Anyhow, I'd have no problem with having the logs of my transactions inspected if it tripped a flag. Personally, I'd even be okay with having all our transactions be public I'd even be ok with selling the right to look over my shoulder (in H$). You could rent your actions to another player for a percentage of their take.

I'll try.

The percentage under the Currently Worth value is what the percentage for that option will become, after you cash out. Not important unless you are holding more than 1 bet on that option (or another option). If you are holding a second bet on that question, it will be affected by the change of value caused by you cashing out.

The basic idea, I think, is that there's a pool of money, divided into the various options (yes/no or 1,2,3..., etc). The relative wealth of each option's cash pool determines the current precentage for each option. When you bet, or when you cash in, you are changing the relative size of the cash pools up or down. That's what changes the percentage. Although you may just wager $50 [and then the new percentage for your option will be ( $inOptionPool + $50)/ ($inMarketPool +$50) ], later on it might have a Currently Worth value of $1000 [ say you bought in with $50 at 2% and it's now at 40% -- the "if right, you get" value would be $2500], so if you take out your money, the new percent will be ( $inOptionPool - $1000)/ ($inMarketPool - $1000).

So the reason your three bets have different percentage values underneath them is that they have different Current Values, so removing them from the market has different effects. In other words, unless you are worried about how removing one bet will affect the other bets you are holding in the same question market, you can totally ignore those percentages.

The percent your option is currently trading at is approx.
100 * (CurrentValue / IfRightYouGet)

Try that for all three bets and you'll see they're very close to the same number. The reason not is that H$ is rounded off to the nearest integer (there's no H cents and nothing like H$7.123456 )

Hope this helps.

In the short term, the cash in date should be tied to the suspend date IMHO.

YES! I've just noticed this on all the many Super Tuesday questions that had a suspend date/time of late Monday night. They're suspended but able to be cashed in. At the time of writing that hardly matters, as the first polls close in just over 12 hours from now. But if the cash in button works until the time of settlement (obviously after the result is known), I can get out of all the bad choices I made at much less of a loss once I know just how bad my choices were. Not good!

Maybe we need two suspends, one as is, and an option to either close cash in at the same time, or to set a fixed amount of time after suspending betting where cashing in is still allowed.

But as the event unfolds the market does this anyway (provided there are enough people following the game and watching their bets and the changing odds). Because as an outcome becomes more and more likely, the percent likelihood p goes up as more and more people either cash out of their predictions of other outcomes or buy into the most likely outcome. So p approaches 100 as the event ends, and the payout P for a wager W is determined by
P = (100/p) * W

So if p is 90% you get around 11% profit on your wager, if p=99% you get just over 1% profit ( 10H$ on a bet of 1000H$), and at 99.9% you get a buck for every 1000 you wager. So if enough people follow the game and the betting, staying open right until the last second doesn't unfairly benefit those who come in late.

That said, the cashing out feature could be unfair if it allows people to cash out after the result is closed to betting and known.

There's two ways to look at a HubDub market, both valid. If the betting ends before the game starts, the market is effectively predicting the odds on the game, and by extension the result. If it continues the whole way through, it's (during game time) more like a stock market, in that it predicts the end result by closely following the actual momentum shifts of the game from minute to minute. If you could map what I'm calling the "momentum shifts" of the game, then the graph of the prediction market should duplicate it, only it would occur some seconds or minutes before the real event's graph (at least in theory).

It may seem, if you favour closing bets early, that you are getting ripped off the other way. But you win and lose exactly the same amount no matter what others do. Your win price is fixed at the time of the bet. So it's more a concern about how others are gaining more because they are using the information gathered during the game to make decisions they wouldn't have been able to if the bets closed pre-game. I can appreciate that point of view, even if I'm more of an in-game bettor myself. That's why I think in-game questions should be kept separate from regular questions (allowing an in-game duplicate of a question that closes before the game - the in-game question could be blocked until 5 minutes or so pre-game, and could borrow the odds from the question that is closing to bets - using its predictive ability.)

This reminds me of investopedia.com, a stock market simulator that allows you to practise investing on the NYSE with $100,000 play money. They also compete over leaderboards, but they have different "games" you can subscribe to. Some last 3 months, some go on indefinitely, some are stocks only, some are stocks and options. The point is that they are different ways of playing the same game, each testing different skills. Maybe down the road HubDub will be able to offer something like this, games and leaderboards for different styles of HubDubbing and HubDubbers?

Patrick

ps admins: Feel free to correct any faulty math or reasoning in the above. You can also use any part of it if you wish as the basis for a math/mechanics of predictions chapter in the FAQ.

I like it.

That was really fun. Many thanks and congratulations to Donal for an excellent job done. You not only suspended predictions in good time, but you also settled them very swiftly. The only extremely minor quibble I had was that the MVP question was suspended with 1 second left in the game, but minutes before the announcement.

Was I the only person to have fairly major connection/update slowdowns with the HubDub site during the game (especially during the last 10 minutes)? If not, then that might be a factor to consider for future in-play events (can you say Beijing Olympics :-) ?).

The idea from drdavin that in-play is unfair to early predictors might suggest having in-play markets kept separate from regular questions (ie you could have both at once, with IN-PLAY questions clearly labelled as such). After all, in-play needs to be monitored for suspension when the end time is not known in advance. So these should maybe only be creatable by the Editor or their monitoring delegates, thereby restricting the number of these to occur.

I was going crazy there in the second half, picking and discarding MVP options, who would win, and whether or not the last two minutes would contain the winning play. It was a blast. Like playing a video game.
(I would have loved to be able to filter out the non-game open questions during the game though.)

I just created the question "Who will be leading at half-time in Superbowl XLII?" yesterday. I have it ending at 6:40 EST which is 23 minutes after the 6:17 kickoff. This seemed to me to balance the need to end before the result is a fact with a little bit of in-play wagering. That said, it's still a bit of a kludge.

I think choice 2) is sufficient for now. As for question creation, I'd say that having the question vetted before it becomes available might save some grief. It also might be a good idea to have the initial probabilities set by someone other than the question creator - this would perhaps allow that person to predict sooner than 12 hours later.

A related issue is the timing of the ending of predictions (especially questions without an exact expected closing time/date). I think we will probably need to have some mechanism for immediately suspending such questions when an outcome "breaks". A recent example was the question "Will Wesley Snipes be found guilty of tax fraud?", which closed about an hour after MSNBC announced the result. Perhaps a mechanism that allows a user to request immediate suspension of a question which then tags and warns all subsequent predictors that their bets may be subsequently disallowed, if the judge agrees with the request for suspension, might help.