Prediction markets for internet points?

Link post

Us­ing real money in pre­dic­tion mar­kets is all-but-ille­gal, and deal­ing with pay­ments is a pain. But us­ing fake money in pre­dic­tion mar­kets seems tricky, be­cause by de­fault play­ers have no skin in the game.

Here’s a sim­ple pro­posal that I think might work rea­son­ably well with­out be­ing too hard to try:

  • Create a ser­vice that tracks In­ter­net Points. Any­one can quickly sign up with a Face­book ac­count and is ini­tially given 1000 points. Points are non-trans­ferrable and are per­ma­nently as­so­ci­ated with your Face­book ac­count.

  • Run pre­dic­tion mar­kets in which peo­ple wa­ger points rather than dol­lars. You can ini­tially copy ques­tions and ver­dicts from ex­ist­ing mar­kets and book­ies. (Ideally you can just fork an ex­ist­ing mar­ket im­ple­men­ta­tion.

  • Keep the point track­ing sim­ple and trans­par­ent. Allow any­one to ob­serve any­one’s point to­tal, and the his­tory of bets that led to those to­tals. Avoid is­su­ing new points. Try to cre­ate a sense of sta­bil­ity. No leader­boards, just ob­serv­ing your friends’ (or strangers’) point to­tals. (Maybe al­low peo­ple to de­ac­ti­vate their ac­count to hide their point to­tal page, but if they re­ac­ti­vate they still have their old point to­tal.)

  • Limit max­i­mum ex­po­sure to 50% of net worth. (But peo­ple can “exit” a mar­ket by tak­ing a bet on the other side—bet­ting on a 5 year ques­tion doesn’t nec­es­sar­ily tie up your points in­definitely.) En­courage peo­ple to op­ti­mize their log re­turns rather than ex­pected re­turns.

  • Make it easy to link to pages for par­tic­u­lar mar­kets so peo­ple can see the cur­rent spread on a given ques­tion; make it very easy to quickly cre­ate a new ac­count and bet on that mar­ket (2 clicks—lo­gin with face­book, and al­low ser­vice to ac­cess your pro­file info).

It may be that a close-enough ser­vice already ex­ists, though I’m a bit skep­ti­cal.

If you are at all in­ter­ested in do­ing this, you have my bless­ing, this is a pretty generic idea (similar to sites that have ex­isted in the past, e.g. Fore­sight Ex­change). I’d also be happy to pay a bounty for an im­ple­men­ta­tion I think is rea­son­able, I’d pay at least $1k for some­thing that seems OK and up to $10k for some­thing I think is ac­tu­ally re­ally good.

Why I think this would be good

The bar­ri­ers to en­try in these mar­kets would be tiny com­pared to ex­ist­ing pre­dic­tion mar­kets—within 30 sec­onds of see­ing some­one men­tion a mar­ket on Face­book, I could have cre­ated an ac­count and placed a bet. I would likely bet in this kind of mar­ket if a few of my friends did.

Some (though far from all) peo­ple would take these mar­kets se­ri­ously. For events in the next year or so, I think it’s likely that these mar­kets would quickly con­verge to bet­ter pre­dic­tions than ex­ist­ing poli­ti­cal pre­dic­tion mar­kets—if prices stayed crazy I think there are plenty of peo­ple who would be ex­cited to step in to fix them even if they didn’t take In­ter­net Points very se­ri­ously (af­ter all, they might be worth some rep­u­ta­tion in the fu­ture, the cost is su­per low, and bet­ting on things is fine). Note that peo­ple who step in early to cor­rect mis­pric­ings would be able to exit the mar­ket af­ter san­ity was re­stored, so wouldn’t even need to tie up their points very long, and that they would have an in­cen­tivize to ad­ver­tise the “easy win” to oth­ers af­ter they’d bet in or­der to free up their cap­i­tal.

If In­ter­net Points are sim­ple enough for peo­ple to re­ally un­der­stand, and peo­ple ex­pect them to stick around, I think they could po­ten­tially cap­ture a big chunk of the al­tru­is­tic up­side from medium-stakes pre­dic­tion mar­kets.

If the over­all work­flow is very smooth, I think there is a rea­son­able chance that the site could spread quite rapidly (spread­ing just as well to new users as to peo­ple who have already made an ac­count).

Note that play­ing this game can still be a fine ex­pe­rience even if you’ve lost most of your points—go­ing from 250 to 1000 points is no harder than go­ing from 1000 to 4000, it’s just that the num­bers are smaller.

Fake ac­counts and pri­vate bets

Ideally you’d have ex­actly 1 ac­count per per­son. But cre­at­ing Face­book ac­counts is quite cheap. Since ev­ery ac­count is granted 1000 points, that’s a prob­lem. Some bad things that could hap­pen, and what you might do about it:

  • Some­one could lose some bets, and then make a new ac­count. But then their points won’t be as­so­ci­ated with their iden­tity (and most peo­ple wouldn’t start a new FB ac­count to get more in­ter­net points), so as long as you mostly care about “your” points then this isn’ ta big prob­lem.

  • Some­one could make fake ac­counts and try to take their points. To avoid this, you prob­a­bly want to make it im­pos­si­ble to trans­fer points ex­cept in pub­lic mar­kets—the worst you can do is have some sock pup­pet ac­counts make bad bets and then take the other side, but if we struc­ture mar­kets ap­pro­pri­ately this might be hard to do dis­cretely. We can also make it against the terms of ser­vice and have sim­ple mechanisms to try to de­tect it (e.g. by dis­al­low­ing new FB ac­counts un­til we have a bet­ter sys­tem in place).

  • Some­one could make fake ac­counts and use them to in­fluence mar­kets, if they care about the mar­ket odds (e.g. if they want their preferred poli­ti­cal can­di­date to look like they have a bet­ter chance in the gen­eral elec­tion). This can be partly ad­dressed by sim­ple mea­sures above (like pro­hibit­ing new FB ac­counts).

It would be nice to fix these prob­lems in a more ro­bust way, so that mar­ket odds are more trust­wor­thy, we can al­low pri­vate bets, etc. That seems like a big pro­ject, over the long run I think the prin­ci­pled solu­tion is ba­si­cally a credit net­work: I in­di­cate that I’m will­ing to trust my friends for IOUs, and then all bets need to be im­ple­mented as a string of IOU ex­changes. This means that the so­cial net­work can even­tu­ally frag­ment into differ­ent pieces (with the fake ac­counts ba­si­cally liv­ing in their own part of the so­cial net­work, which has very few trust con­nec­tions with real ac­counts), and the mar­ket odds will po­ten­tially be differ­ent for differ­ent peo­ple. That makes ev­ery­thing way more sub­tle to think about, and the im­ple­men­ta­tion seems way more com­plex, but it seems to me like it should work over the long run and is where you’d re­ally like to be.

(You could also in­te­grate that sys­tem with iden­tity ver­ifi­ca­tion ser­vices who re­ceive a high de­gree of trust, and then a ser­vice it­self could effec­tively be de­ac­ti­vated—have all of its trust ex­hausted—if it cer­tified a bunch of users who lost money.)

Subsidies

A ba­sic prob­lem with this mechanism is that there is no sub­sidy (ex­cept for noise traders). It’s not as bad as ex­ist­ing pre­dic­tion mar­kets, where the house takes a huge cut, but not as good as most “real” ap­pli­ca­tions of pre­dic­tion mar­kets where some­one in­ter­ested in the info is will­ing to pay for it.

If I’m bet­ting with some­one rea­son­able, at least one of us is los­ing ex­pected log-points (since the log bakes in risk aver­sion). So the mar­ket can’t work with par­ti­ci­pants who trust each other’s ra­tio­nal­ity, un­less peo­ple just love gam­bling.

The most nat­u­ral way to fix this is to sub­si­dize mar­kets. But do­ing this makes the gov­er­nance and in­ter­pre­ta­tion of In­ter­net Points harder—does some­one with 4000 points have a great his­tory of los­ing bets, or did they just col­lect a sub­sidy? who are we effec­tively sub­si­diz­ing, and how do we make that call?—so I’d pre­fer avoid it. If nec­es­sary, prob­a­bly the nicest way to do it would be for the house to provide a mod­est amount of liquidity at the prices im­plied by ex­ist­ing mar­kets. Ex­ist­ing pre­dic­tion mar­kets are suffi­ciently ir­ra­tional that I think this would provide a sig­nifi­cant in­cen­tive to par­ti­ci­pate.

If prices on In­ter­net Point mar­kets are used se­ri­ously as fore­casts, it also in­tro­duces an ad­di­tional set of in­cen­tives—some peo­ple want to ma­nipu­late fore­casts (e.g. to make it look like their preferred can­di­date would be more likely to win the gen­eral elec­tion). So the av­er­age truth-seek­ing trader ex­pects to get money by bet­ting against them, and this can make the whole sys­tem work.

In some sense this is ac­tu­ally the only case where you re­ally needed pre­dic­tion mar­kets any­way—if ev­ery­one is be­ing rea­son­able then we can just talk it out and share our pre­dic­tions, and that naive mechanism only re­ally breaks down when some par­ti­ci­pants be­lieve other par­ti­ci­pants are be­ing pre­dictably un­rea­son­able (for what­ever rea­son). For bet­ter or worse, this is a pretty typ­i­cal situ­a­tion.