The Markets-for-Markets Competition, One Week In
Welcome back to my discussion of my markets-for-markets competition. If you're just joining us, you can read more about this competition in my previous post:
To quickly summarize, I am running a competition on Manifold Markets to see who can create the best market to relate upcoming events to future events.
The contest has four entrants, each of which has been credited with 2000 mana to go towards liquidizing the market and boosting the markets in user’s feeds. In this post, I’ll analyze these markets and bet on them.
Entrant 1: Bitcoin ETFs and new all-time-highs
The first entrant comes from manifold user Evan. This market resolves based on the outcome of two events: Whether or not the SEC approves a spot Bitcoin ETF in 2023, and whether or not Bitcoin goes on to reach a new all-time-high price (in USD terms) by the end of 2024. You can see the market here:
And here you can see a visualization of the correlation of the outcomes (as of time of writing), using Johnny Ten-Numbers’s correlation tool.
My Analysis
I am not a financial professional (this is not advice). However, my understanding is that a Bitcoin ETF, if approved, would allow a larger number of individuals or institutions to effectively buy Bitcoin by making it more convenient to do so. It is therefore thought that if the ETF is approved, this will lead to increased demand, which will then raise the price, making it more likely that bitcoin will attain the all-time high.
How high would we expect the price to rise? In late October, news source CoinTelegraph made a tweet saying that such an ETF had been approved. This report later turned out to be mistaken. Nevertheless, the bitcoin price spiked from around $28000 to about $30000 in the space of a few minutes (incidentally, this caused some excitement on this market).
Is it reasonable to think that $2000 is a sensible amount for the Bitcoin price to increase with this kind of news? Would it have gone even higher if the CoinTelegraph story were not debunked sooner? Should we consider the people trading on this news were just making a bad decision and discount the spike entirely? These are all questions about which I have a lot of uncertainty, but seeing as how the exact timing of these approvals might not be so important, I’ll take $2000 as a rough estimate of the price increase.
If we look at Deribit Deltas to get a sense of the likelihood of Bitcoin price outcomes, we see that the market thinks there is about a 0.30 delta for Bitcoin being above the all-time high price of 65,000 USD in September 2024, and a 0.27 delta for it being above 70,000 USD (the all time high is 67,617 USD). This suggests to me that the impact of a 2,000 price increase is maybe around 1.5%.
My response is therefore as follows. I will buy the probability of attaining the high up in general, but I will also buy the conditional probabilities closer together.
Entrant 2: Gemini and GPT-5
Our second entrant is this market by Infohazard Assessor, which resolves off of the release dates of the forthcoming Gemini and GPT-5 Large Language Models (LLMs) from Google and OpenAI respectively.
Correlation is nearly nonexistent as of time of writing:
My Analysis
As the creator commented, this market is an attempt to measure whether there is an “arms race” in LLMs. This refers to the concept that large AI labs, being in competition with each other, will try to move faster in creating AIs if they perceive that they are falling behind.
While I see this perspective, it is not clear to me whether it is correct, or whether this question captures it fully. It might be the case that AI labs are just building models as fast as they can, because regardless of what other labs do, they are economically incentivized to get to market as fast as possible. Perhaps its even the case that if Gemini comes out sooner than OpenAI expects, OpenAI might even spend longer on GPT-5, on the concept that they will want to differentiate their product by quality.
In any case, I do not know much about this subject, so I am inclined to defer to the market. I will not make any particular bets, except to arbitrage with the underlying binary markets.
Entrant 3: Starship and Moon Landings
Our third entrant comes from user James Calder Knight. It links Starship reaching space in 2023 to a human landing on the moon by 2028.
Correlation:
My Analysis
SpaceX seems like they will soon get approval to launch their Starship for the second time in the next few weeks. The last time they launched was in April, around 7 months ago. If the launch fails, how long will it take for them to try a third time? Presumably they are in the rhythm now and it will take less than 7 months - it’s hard to tell if the limiting factor on this was their modifications to the ship or just the FAA approval. It seems they have made some major redesigns, but perhaps they only did this because they knew the FAA would take their time. Still, it would be surprising if they cut down their turn-around time by more than an order of magnitude. Splitting the difference, I figure that if they fail this time, they’ll try again in 3 months or so.
At that point, they might fail a third time. I’m not sure how embarrassing this would be, but base rates say that it is harder to get a Starship to space than to explode it on the way to space. Hopefully SpaceX is not fooling themselves about the possibility of this working at all. I’ll round back up and call the total expected delay in getting to space if they fail now relative to if they succeed now an even 7 months.
How likely is this to impact the next human moon landing? NASA plans to land a human on the moon with its Artemis program, for which it has commissioned a Starship-based Human landing system. It’s reasonable to think that delays in Starship getting to space now map one-to-one onto delays in Starship being ready to land humans on the moon. Is SpaceX the only bottleneck here? Musk says “no”, but he has a reputation for bluster when it comes to his companies.
So if we assume a 7 month delay if the 2023 launch fails, the question is: “What is the likelihood that the Artemis human landing mission takes place in the 7 month window around the end of 2028?”. This market currently gives a 28% chance of getting to the moon in 2028 overall, which would make for about a 16 percentage-point difference in probability for a 7 month window.
The market currently gives a difference of only 6 percentage points, so I’ll buy to increase this.
Entrant 4: Israel-Hezbollah and US-Iran conflict
Our last entrant comes from Jack, and deals with whether there will be a 400 death conflict between Israel and Hezbollah in 2023, along with whether the US and Iran will be at war before 2025.
Correlation is 0.26:
My Analysis
This is another subject about which I know little. Here is a news piece on the potential for Hezbollah to join the current conflict in Israel. My takeaways from this are:
Hezbollah is a fighting force that is more capable of opposing Israel than Hamas.
Nevertheless, the Israel Defense Force is still more militarily capable than Hezbollah.
Hezbollah has ties to Iran.
What does this mean for this market? The pro-correlation argument probably goes something like: “There is a possibility for Hezbollah to attack Israel, and if it does, Iran and the US could be drawn in WWI-style”. This is presumably why the market thinks it’s almost impossible that the US and Iran could come to blows without Hezbollah moving in. It’s hard for me to assess the numerical likelihood of these outcomes though - it seems like a complicated situation with many actors, each with many potential motivations. I will buy 5 or so shares in NO-NO on the general principle that that outcome is trading at less than 1% and the situation is uncertain.
Who is Winning?
As described in the original post, the outcome of the tournament is to be determined by the time-averaged mutual information. Here are the current standings (not time-averaged, I haven’t written that code yet).
The Middle East:
Individual probabilities are [[0.07, 0.38], [0.014, 0.54]].
Mutual Information: 0.028
Space:
Individual probabilities are [[0.19, 0.33], [0.11, 0.37]].
Mutual Information: 0.011
Bitcoin:
Individual probabilities are [[0.24, 0.26], [0.21, 0.29]].
Mutual Information: 0.00181
LLMs:
Individual probabilities are [[0.29, 0.27], [0.20, 0.23]].
Mutual Information: 0.00136
Congrats to Jack, who seems to be in the lead. I’m actually pretty surprised by this ordering — Frankly, my guess for how this would have played out would have been almost the exact reverse. My conclusions are as follows:
For some topics that seem correlated at first blush, you can analyze them relatively thoroughly and end up concluding that there is very little correlation.
Even events far in the future can potentially be impacted by the near term.
Finding an event which is a hard prerequisite for a particular outcome, even though it in theory gives up on the maxing out MI, can still be very powerful.
There is still probably a lot of mana on the table here.
Good luck again to all the contestants!