TL;DR: write an arb bot that trades the markets linked in this post against their underlying markets in at-least-slightly non-trivial ways, post source code and some info, get bounty.

There are a bunch of problems with high-certainty markets having the correct prices. They seem to show up when odds are more extreme than 90%, time until resolution is nontrivial, and/or when liquidity is low. Frequently, correcting these probabilities just isn't worth people's time, effort, capital / opportunity costs, or risk of bad resolution.

However, sometimes we can get market reports that this is going on. For example, this market currently expects 85% chance that all of its underlying markets will resolve in the "expected" direction:
/EvanDaniel/long-shotish-bets-how-many-of-these

But if you took the prices of the underlying markets at face value and assumed no correlation, you'd expect about a 67% chance of that happening.

Or, on a shorter timeframe, the list of markets about nuclear tests and detonations in 2023 has some arb opportunities (such as Russia conducting a test by EOY being more likely than any detonation by EOY). It also looks like the market somehow expects either everyone to test something or no one to test something. But doing a purely mechanical arbitrage to correct that is impossible, without a market like this one where you can hedge the correlation:
/EvanDaniel/nuclear-tests-and-detonations-2023

Before fees and slippage, you could make a guaranteed mechanical arbitrage with about 1% return by EOY. Factor in loans and so on and the annualized rates of return are... well, not great, but also pretty good by the standards of the ordinary financial world and risk-free returns. (Well, risk-free until Manifold goes under, or floats Mana, or whatever, but you get the idea.)

The markets:

The ones listed above, plus also all the other derivative markets on this topic.

Feel free to suggest more markets and I'll be happy to include them. Create more that follow the same format and I'll award a bounty to at least cover the creation costs on the first couple submissions.

At the moment I believe the format should be:

• A linked single-outcome multiple choice market

• Resolves to count of markets that resolve "as expected". (So if there are a mix of > 90% markets and < 10% markets in the underlying list, it should not be a count of "resolves Yes". Inverted count of "resolves unexpected" is also fine if that improves readability.)

• Connects more than two underlying markets, so that it isn't just a joint market.

The main bounty:

Write a bot, or modify an existing bot, to arbitrage some of these markets. The bot should, at a minimum:

• Be open source (ish): code should be available such that someone else could run a version of it. If it uses config files where you're running a different version or "secret stat-arb / fancy correlation estimation function goes here", that's fine; the code you're running doesn't have to match the code you share entirely, but the public stuff should be enough to at least theoretically make some profitable trades.

• Have placed some arbitrage trades involving the markets linked above

• At least some of the trades are not-entirely-trivial. For example, right now the 2023 nuclear testing market has 94% of 0 underlying markets resolving Yes, but the Russia test market is at 7%. Make a trade at least one tiny step more complicated than that.

• Write at least a few sentences about it.

M1500 for the first bot to do this. Another M1500 for the first bot that wasn't already public before this was posted (could be the same award as the first one). M1000 for the next entrant, M750 for subsequent entries.

Additional lesser amounts per my judgment (probably M250-500) for:

• Your bot at least sometimes leaves limit orders standing for human-trader-relevant amounts of time.

• Does some amount of interesting statistical arb work.

• Has a managram interface where I can loan it working capital and receive interest. (I will also do this if the rates look sane; they do not need to be good compared to my opportunity costs.)

• Suggesting (or creating) new markets that your bot can use to be more effective.

• Your bot also trades in any of the Joint Markets Contest markets.

• Trades in /1941159478/will-republicans-win-pennsylvania-g, /HanchiSun/etf-china, or similar markets.

• Trades in the dependent markets of the "36 weird markets": /Ernie/2024-biden-wins-which-states-multip

This bounty is currently severely underfunded. If the funding runs out before everything has been awarded, further submissions might not get any awards. But if we find ourselves there, hopefully that means there's enough interest that I'm excited about continuing it and adding more funding.

Obviously funding contributions are welcome. But I'm also writing this project up for the Manifold Community Fund. If it finds funding there (and even if it doesn't, if there seems to be some general interest), I'll increase the funding. Increased funding will be used to raise the bounty amounts, increase the number of bounties paid out, and pay additional bounties to already-awarded entrants to match the increase in bounty values.