It seems to me that there are three popular narratives running at the same time on Manifold:
1. There is a >9% chance the democratic party strategically retires Biden and brokers a different candidate
2. If Biden is not the candidate, Harris is likely the candidate
3. Harris is a very poor candidate with a low chance of success

Personally I find it very unlikely that all three of these could be true at the same time. And personally, I think #1 is the false one

@Tumbles Yeah, I agree with this, which is why I still have so many YES shares on the Biden nomination market. If Dems could go back in time and make Biden drop out in early 2023, or better yet, not run in the first place, and then replace him with someone like Whitmer, maybe they would, but it would be very difficult to do it at this point. Also, choosing the nominee through a brokered convention rather than a democratic primary process is a really bad look, even if they can say, "Filing deadlines forced us to do it that way."

I don't think this market is any better than https://manifold.markets/ManifoldPolitics/who-will-win-the-2024-democratic-pr-47576e90fa38. Most of us have concluded that conditional markets (like this) provide no better prediction accuracy than non-conditional markets (like the linked one). The strength of incentives to predict/price accurately is the same, and N/A creates further unfortunate complications (such as making it much harder to arbitrage with other markets).

@jack I generally agree, but part of the purpose of this market was to get another data point in testing that.

@jack Also, you get one more decimal place, right?

@Irigi I claim that precision is mostly noise, not real precision.

@jack You might be right, but I am not that sure. People can have quite good intuition about the ratios between different outcomes, if not on exact probability of each. (And might not care to put too much mana to push outcome from 3% to 1%, while they would easily spend few mana to move 30% to 10%.)

@Irigi Yes, I agree that it's often much easier to reason about conditionals and chains of probabilistic reasoning - "A) Will Biden be the nominee? B) If not, then who?". But for markets where the conditional is very unlikely, we have seen in many markets that the probabilities are sometimes ok while sometimes way off, because people don't have much incentive to correct them (since any trades will most likely be N/Aed).

But ideally Manifold supports this by showing a conditional view ("If Biden is not the nominee, then who will it be?") over a market that is implemented non-conditionally ("Who will be the nominee?"), and letting you bet on the conditional view. This idea has been discussed a bunch and I think it's moderately complicated, so I don't think it'll be implemented anytime soon, but conceptually I think it's an obvious improvement.

(And might not care to put too much mana to push outcome from 3% to 1%, while they would easily spend few mana to move 30% to 10%.)

I agree, and I think this is kinda exploiting irrationality. I think it is true that people are (somewhat) more willing to move it from 30% to 10%. But rationally, they shouldn't be - the expected profit is the same. It's good if you want more predictions on this specific market, and bad in the sense of distorting people's attention towards markets that have less real value (as measured in mana) than other markets.

@jack There are aspects beyond mean value maximization. Namely:

• Balancing risk of single point of failure (Kelly bet)

• Keeping enough free mana to wait for new opportunities (unclear to me how to formalize, but I think there probably is some optimum ratio of mana to keep free)

And maybe others. I am not saying you are wrong, but I am not convinced you are right either. This is quite complex optimization problem.

@Irigi Expected exposures are exactly the same, so EV, Kelly, etc are all identical.

I think the conditional market either requires more capital or the same amount of capital, for the same expected payout. (Edit: I previously wrote something about the difference in capital requirements that was wrong, will try to fix it and repost)

@jack With Kelly, I meant it like this: Say we have regular AMM R and conditional AMM C. Both have liquidity pool 100 M, and the trader has 1000 M. The regular market is at 3% and belief of the trader is at 1%. The conditional market 90% resolves N/A, and otherwise is at 30% and the trader has belief 10% about the conditional outcome. They both should have the same mean value, right? (0.1 * (0.3-0.1) = 0.03-0.01). We can calculate the Kelly exactly like this:

Profit of R if invested M mana if NO: n - n (y/(M + y))^(p/(1 - p)) := AA

Loss of R if invested M mana if YES: -M

Kelly Criterion: maximum of (1-p2) Ln[1+AA/1000]+p2 Ln[1-M/1000]

Here y = n = 100 (because liquidity pool is 100 M), p = 0.03, p2 = 0.01. Numerically, the maximum lies at 151.714 M, which moves R to 1.18%. (Ideal would be 1% if there is infinite mana).

For the conditional market, there will be y = n = 100 (same reason), p = 0.3, p2 = 0.1, but Kelly will be

Kelly Criterion: maximum of 0.1*(1-p2) Ln[1+AA/1000]+0.1*p2 Ln[1-M/1000] + 0.9*Ln[1]

Numerical maximization leads to investment of 128.829 M, which moves the conditional market to 11.61%.

The fact that you should bet smaller amount in the conditional market is quite counter-intuitive to me. The conditional market ends up with better probability (closer to trader belief). But the difference is tiny. (in this case)

Curious to see how this compares to this new market, which does not resolve to N/A if Biden is not replaced but instead resolves to the 2028 dem nominee:

Given that Harris has been at about 50% in this market but is only at 9% in the biggest 2028 nominee market, we should expect her to end up somewhere in the middle for this new market. Exactly where in the middle will depend on how likely Manifold believes it to be that Biden is actually replaced.

oops ignore this

typo

@TiredCliche fixed!