Remedying Incentive Misalignment when Tips dwarf Stakes

Before I start here is some essential background knowledge

  1. Content creators must always stake money on their writing (skin in the game) between a set minimum and a set maximum.
  2. Markets evaluate the quality of the contribution.
  3. Reputation scores are updated in accordance with market scores.
  4. Aside from stake payouts, tips are expected to be a large revenue stream for content creators. However tip payouts must be tethered to market performance as best as possible.
  5. Olas employs various anti-collusion mechanisms such as MACI, pairwise betting and more to make collusion to influence these markets extremely difficult.

The starting point for all Olas information markets is that the journalist must stake some money on the article they are publishing (the long side). The short side of the market is the Olas Global Pool. The writer is competing with it for a payout both from his/her stake and from tips given to the article.

Dealing with the stake is easy. If they stake $100 and had 30% of the market agree with them they’d only get $30 back. However the tipping pool payouts create complications for the integrity of the incentive system. If for example this particular article went viral and garnered $1000 in tips (with 30% of them positive), the writer would make a total profit of $330 ($300 + $30 returned from stake) from their $100 stake. That’s 3.3x profit despite receiving a poor score. Although there would be reputational damage, there is a strong incentive to push out low quality content that engenders a reaction and goes viral. In other words we’d just have recreated the incentives that make the legacy media system so bad!

So we need a mechanism to better tether payouts from the pool of tips to the stake. We propose an exponential payout curve with a straight line between 45%-55% market share for to represent a wide confidence interval for indecisive markets for occasions when the tipping pool (T) > stake (S). The base curve is T/S = 1. Payout outcomes are particularly decisive below 30% and above 70% meaning low scoring articles are severely punished and high scoring articles are well-rewarded. As you can see the S is not symmetrical - it’s steeper below 45% - because the risks aren’t symmetrical either. Our main concern is to guard against high engagement, low quality information. So when T > S, it’ll payout far less than a linear payout for poor articles. The mechanism is further tethered to stake size with a stake power factor and a low stake penalty factor further ensures low stakes for high engagement articles aren’t particularly profitable for outcomes below 45% market share.

This is what creates the asymmetry. If an article goes viral and T is 100x the size of S, there would still be profit at a 30% market score but not a huge amount and the reputational damage would be severe and affect future income. Conversely, the payout to the Olas Pool to fund future quality writing would be substantial. So overall we believe this exponential tip payout mechanism achieves our goals quite well.

However, this solution that protects against bad incentives for poor information introduces an undesirable consequence for good articles when T dwarfs S. In the case that an article scores very well and goes viral, the journalist might receive only half the tips or less if their stake was small. Star performers that actually bet the maximum stake on their work yet have large followings would be particularly badly affected as even though they bet as much as they could and scored very well, they’d be a victim of their own success with tips received far larger than their stake.

To mitigate this issue, we propose two further adjustments: A high performance multiplier and a reputation multiplier. The former by definition would only apply to the ride side of the curve (as that’s where the high scores exist), whereas the latter would apply along the entire curve. Those that stake a lot, score highly and have a good reputation from a strong track record, will earn a large majority of the tips available for the article. This is naturally a desirable outcome and a healthy incentive to perform.

Given the likelihood that T will often be larger than S, this extremely punitive and rewarding payout mechanism (depending on what end of the spectrum an article lies) ensures healthy incentives for accuracy and good writing when the journalist’s skin in the game is dwarfed by the possible payout available.

In this vein we get to embrace tipping, our big bet on reviving the fortunes of media in the face of dwindling ad and subs revenues, while maintaining the intergrity of our quality control systems.

Feedback welcome!

5 Likes

The reputation and high-performance multipliers could have a big impact on the long-term success of content creators.

Established creators with strong reputations will likely benefit more from these multipliers, earning higher rewards and building greater trust with their audience.

However, newer creators might face challenges in gaining traction without an initial strong reputation, which could slow their growth.

It would be interesting to know how these multipliers balance rewarding experienced creators while still allowing fresh voices to emerge and succeed. Is there a plan to support new creators in the early stages?

2 Likes

I like how the mechanism should filter low-quality content and focus on incentives for credibility and not just hype. That can also inspire contributors to have a kind of longer-term strategy for staying on Olas and building their reputation.

…And this would be the result of that strategy, where they can boost their earnings by focusing on contributing relevant and correct content. Incentivizing quality is always a good idea.

On the flip side, how can readers be sure that the contributors won’t place ‘’safe bets’’ with their content? What if writers focus on expressing their not-so-bold or innovative opinions on burning topics just to avoid reputational damage?

We have a Bayesian Truth Serum mechanism that’ll punish safe bets that are wrong. For safe bets that are right, they’ll be rewarded as right is right. However if it’s simply pointing out banal truths I can’t imagine they’d earn much from tips. People won’t tip a Captain Obvious that writes something most people have long known to be true.

2 Likes

This raises some intriguing points about balancing incentives in content creation, especially with the complexities introduced by tipping and staking. The proposed exponential payout mechanism seems like a thoughtful approach to minimize the potential for low-quality content to thrive, while still rewarding high-quality work.

However, I wonder how this system will be received by creators. Will the risk of reputational damage deter writers from taking bold stances, or could it stifle creativity? Additionally, how do you plan to educate both creators and the audience about this new model to ensure its success? It’s a fascinating concept, and I’m curious about how you envision its implementation in practice!