Alex Tabarrok advocates for crowdfunding systems with *Refund Bonuses*. I think this might be a natural occurrence of a money pump against Causal Decision Theory pledgers
post by mako yass (MakoYass) · 2022-03-14T07:27:06.955Z · LW · GW · 13 commentsThis is a link post for https://marginalrevolution.com/marginalrevolution/2021/06/better-crowdfunding.html
Contents
13 comments
Essentially, his version of Kickstarter has Refund Bonuses. His rationale is that there's little benefit to investigate or promote a probably-failing crowdfunding project, because that costs you time, so that could put projects in a bad equilibrium where no one will strive to believe in them tomorrow because nobody believes in them today. So Refund Bonuses breaks the vicious cycle by compensating that time-cost, by giving out a bonus to anyone who pledged, if and only if the project fails to meet its funding threshold. A paraphrasing of it is: It sends a strong signal that the project is confident that it will succeed, so if you would have been doubtful, you shouldn't be now.
And, for a worthy project, the refund bonus would rarely need to actually be paid, so it costs little to offer them.
Hearing about this, something occurred to me. I think this might be a natural dutch-book against (trap, for) CDT (Causal Decision Theory. Background: https://arbital.com/p/logical_dt/ ).
The vice of a CDT agent is that it sometimes cannot anticipate that the decisions of other CDT agents will be entangled with its own decisions.
It'll go like this: An opportunistic CDTer might decide to look for campaigns that will fail, so that they can extract refund rewards. They watch the market for a bit, develop a sense of the patterns of failure and convince themselves of a strategy.
A bunch of other CDTers do the same thing and develop mostly the same strategies.
This turns into a game of chicken.
But CDT consistently underestimates the correlation of its decisions with those of other CDT agents, because it cannot use its own decision as evidence about others' decisions.
I'm not completely sure how much CDT underestimates by.
But if it's underestimating by a lot, many of its projected failing campaigns will be driven by cynical pledges unto success. This would run against my usual intuitions about CDT... Usually, it would underfund public goods, it appears to me that refund bonuses would lead it to overfund public goods. I guess the common factor is that in both cases it's failing a collective action problem.
Very few of the CDTers would end up making money by fishing for failure payouts. The live players in this game admit that the CDT way of thinking has serious issues and they stop using it in this context (They might not know the LDT angle, so they'll just say things like "It's a fool's game. It's gambling." and at their best "This market is entirely anti-inductive."), but a dead player, who has no meta-rationality, may trudge on, always expecting to score, never seeing the invisible acausal logical threads connecting it to the mechanism of the trap.
13 comments
Comments sorted by top scores.
comment by Yoav Ravid · 2022-03-15T17:30:18.658Z · LW(p) · GW(p)
I'm confused. The baseline and PE20 results from the chart in the post don't match the ones from the table in the paper. What am I supposed to make of this? (also, P20 doesn't even appear in the chart)
↑ comment by Yoav Ravid · 2022-03-16T16:33:27.248Z · LW(p) · GW(p)
I contacted Alex to ask about this, here's his response "thanks for catching that. My fault. It can take years to publish a paper in econ so my slides came from an earlier version. We were asked to add some more experiments in a revision and so the numbers changed. The published version is definitive."
So the pretty chart in the post should have the baseline at 44%, PE20 at 67% (higher than the current highest) and another column for P20 at 61%.
I made a new chart for my post on this suggestion [LW · GW].
comment by Yoav Ravid · 2022-03-15T08:08:10.623Z · LW(p) · GW(p)
Is it the person who seeks funding who has to pay the refunds? If so it means they must make an investment before starting the project in order to be able to ask for funds. But If it's the platform paying out the refunds it introduces a different problem, where someone can profit by starting a project no one would want to fund, let a friend contribute some but not enough to fund it, and then split the refund money with that friend.
Either way, what happens with project that are harmful, rather than beneficial (as we can assume most project to be)? In such a system it seems you're not only incentivized to support beneficial projects, but also harmful ones, in the hope that they won't get fully funded and you'll get a refund. Will that cause more harmful project to succeed? Or will the "pivotal backers" never actually back the project and it would still fail? In that case (and if refunds come out of the campaigner's pocket) it would actually disincentivize harmful project by making campaigners pay for them (yet will still reward early backers for backing them).
Overall I this idea is very interesting and I think I like it. I'd love to see a crowdfunding platform experiment with it.
Replies from: MakoYass↑ comment by mako yass (MakoYass) · 2022-03-15T21:38:01.307Z · LW(p) · GW(p)
Will that cause more harmful project to succeed?
In reality I'm not sure the trap would remain effective for long enough for too many of those to start turning up. Humans aren't rigidly CDTish. They'll catch on. Perhaps many professional traders already have some principle against playing games of collective chicken.
I guess a good question here is... is the opportunity cost of assessing the failure rate, to the level of accuracy where the risk of project success is low enough that you can be sure that you'll get your refund bonus, actually lower than the refund bonus. I think it has to be proportionate to it! I'd guess that earnest pledger's work, figuring out whether the project is good, would be roughly equally to the exploitative pledger's work of figuring out whether the project will succeed. But I think the answer is no. For the exploitative pledger, success is a risk, for the earnest pledger, success is a boon, the earnest pledger doesn't need the refund bonus to be so high to justify their time investment because they don't have that risk to offset, so the refund bonus will be too low to justify the work for most exploitative pledgers.
So in most cases the refund bonus extraction game wont be profitable. But that doesn't mean it'll be unprofitable in all cases. Sometimes the analysis will be especially easy (the product will be clearly useless). And sometimes one extractor's analysis will leak, and others will imitate them (creating a very dangerous game for them all), or, oh, one extractor will pledge multiple times with alts ..
comment by Pattern · 2022-03-14T21:48:18.499Z · LW(p) · GW(p)
Who pays the cost for refund bonuses?
Also, that money could also be used for funding public goods so - expecting them to become overfunded (when they are losing a source of funding) seems to miss the mark. 'The money is more poorly allocated' seems like a bigger issue, that could happen.
The alternative is refund bonuses coming from the platform (kickstarter) which is already taking a cut.
comment by Gunnar_Zarncke · 2022-03-14T11:14:13.503Z · LW(p) · GW(p)
Seems like a case where civilization should offer something to improve overall coordination - not the project itself which can't (where is the money for the bonus supposed to come from?).
Or maybe have a prediction market of these projects. Maybe let the project itself subsidize it - could be cheaper than a bonus.
comment by Alexander Gietelink Oldenziel (alexander-gietelink-oldenziel) · 2022-03-14T10:57:06.558Z · LW(p) · GW(p)
A very striking example is the Art - and now the NFT market. Art tokens have (almost) no use value, yet they can command enormous prices. Their value is completely derived from some sort of collective/common knowledge game & coordination/correlation of many players (or few players with a lot of weight).
Naively, one would expect art to have no value - and perhaps a CDT agent in your sense (I don't quite get this) would never enter the art market. Ofc real fortunes have been made by shrewd operators so empirically there is something there.
Replies from: MakoYass↑ comment by mako yass (MakoYass) · 2022-03-14T22:14:22.062Z · LW(p) · GW(p)
Mm, to add context, you're mentioning this because it's a very anti-inductive market, yes? And yet people keep participating. So why wouldn't they keep participating in the refund bonus extraction game of chicken.
comment by Pattern · 2022-03-14T21:50:23.762Z · LW(p) · GW(p)
This article is wrong - starting with the title.
In 1998, I designed the “dominant assurance contract” (DAC) mechanism for producing public goods privately. In my latest paper, just published in GEB written with the excellent Tim Cason and Robertas Zubrickas we test the theory in the lab and…it works! Kickstarter hadn’t yet been created when I first wrote but the DAC mechanism can now be easily explained as a Kickstarter contract with refund bonuses. On Kickstarter and other crowdfunding sites you contribute to a project and if a contribution threshold isn’t reached you get your money back. The Kickstarter contract is useful but it’s still easy for a good project to fail because there are many equilibria with non-funding. For example, if I think that you won’t contribute then I may decide not to contribute and if I don’t contribute then you may decide not to contribute.
It also answers some questions, like 'who might pay for the bonuses'?
Third, refund bonuses pay for themselves! In theory, refund bonuses are never paid but in practice, as we have seen, some socially valuable projects fail even with refund bonuses. Nevertheless, for reasonable markups it’s still in an entrepreneur’s interest to use refund bonuses because the greater success rate more than pays for having to pay modest refund bonuses when a project fails.
We think refund bonuses can substantially improve crowdfunding and we hope to partner with a crowdfunding site to run a field experiment. Contact me if interested!
Read the whole thing.
The paper has more to say about that:
Moreover, we find that even taking into account campaign failures, refund bonuses can be financially self-sustainable suggesting the real world value of extending assurance contracts with refund bonuses.Replies from: MakoYass
↑ comment by mako yass (MakoYass) · 2022-03-14T22:06:08.044Z · LW(p) · GW(p)
What's do you think is wrong? I don't see any contradictions here.
Are you confused about who I'm saying is getting dutch booked? I'm saying pledgers dutchbook themselves, the project will be more than fine, it would be extremely good for the project. It seems like a very good mechanism from the project's perspective, and I approve of it.
Replies from: Pattern↑ comment by Pattern · 2022-03-14T22:26:12.682Z · LW(p) · GW(p)
"Alex Tabarrok proposed improving crowdfunding mechanisms with Refund Bonuses."
The proposal predates kickstarter.
Replies from: MakoYass↑ comment by mako yass (MakoYass) · 2022-03-14T22:33:33.528Z · LW(p) · GW(p)
I see how that can be misleading. I'll try to clarify it. The reason it ended up looking like that was that "kickstarter with refund bonuses added" is, as he acknowledges, a really good way of describing it, even though it was not a product of taking kickstarter and adding refund bonuses.
Replies from: MakoYass↑ comment by mako yass (MakoYass) · 2022-03-14T23:06:33.432Z · LW(p) · GW(p)
For posterity, the original title was
Alex Tabarrok proposed improving crowdfunding mechanisms with Refund Bonuses. I think this might be a natural occurrence of a dutch book against Causal Decision Theory
I also removed these sections which I kind of left in by accident and had already decided at the time of posting that I couldn't really stand behind. Sorry about those.
Could be true, but I think I was probably understating the value of the credible signal that is sent by having refund bonuses, even for a LDT agent.
As a Logical Decision Theory (LDT) sort of decisionmaker, wow, I think that might all be predicated on a level of cynicism that I have difficulty relating to. When I'm considering a crowdfunding project, I'll generally infer the existence of a group of faithful, likeminded actors whose decisions are entangled with mine, who would be happy to investigate and promote the project while it is small, knowing that many others will tacitly move along in train with them, and if that group was big enough that the project might succeed, we would go ahead and investigate and promote the project, and that would usually be enough, I think.
Causal Decision Theorists (CDT) floating around, who can't coordinate like that, or — more charitably — maybe there are people who just can't accurately infer the size of their good faith entangled cohort very well, and they underestimate it, so they can't justify adhering. And maybe we need them on board. For them, Refund Bonuses excites me.