Exploring Prediction Markets and Equity Research - Week 4
Intro:
A little bit of a disruption in the article cadence as I just finished up a move, my apologies. On the bright side our internal predictions testing is going nicely and the skeleton is working. Still needs some fleshing out on our end, but soon we’ll have some self-hosted questions up on our site. Hopefully before Manifest 2025 which we’ll be attending.
We’ll also be resuming car market related posts this week with some updates in the pipeline to our KMX coverage. Additionally our Twitter will soon have daily updates on our coverage, which we’ll continue to work to expand.
For continued updates, feel free to subscribe.
Navigating Uncertainty:
Not many market updates, so let’s do some monologuing.
Writing about markets is cool because they never cease being interesting! Coreweave and United Healthcare have seen extreme volatility over the past week. What’s fun is that when things are that volatile, nobody has a clue what’s going on. You think you’re knife catching UNH and then suddenly there’s a WSJ piece about ongoing criminal investigation and the CEO steps down? Then the company releases a statement that the WSJ piece is fake news?
The outcomes there get super interesting because
The information environment is so poisoned nobody has any idea what to believe
The potential news causing the uncertainty may have high impact when resolved
The stock is trading on so many news pieces so rapidly it’s impossible to know consensus
Sell-side and 8-K’s aren’t designed for rapidly shifting narrative environments.
The result is a stock with a few hundred billion market cap being more volatile than a shitcoin.
The typical investor of course simply avoids such situations as getting up to spin on the US healthcare system in a couple of days is a herculean task. For those interested in differentiated performance however, avoiding the actions of the typical investor is the name of the game (ideally to the upside).
So how does one do that?
Being really smart and making good guesses isn’t necessarily scalable, so I believe we’ll likely see some emergent tools start to pop up over the next few years to address this problem. When trying to solve for the future, I like to think about things at the extremes. Like if I had a gun to my head and had to choose between never shopping at Amazon or in person retail again, I’d choose Amazon. I’d like to think that’s a good insight, so let’s explore a metaphor for sense-making.
At the core, the hard part about these super volatile names is lots of information must be processed in a short period. For example, say someone walks up to you on the street and offers you 1 ounce of palladium for $100. Ignoring the potential scam of palladium street vendor, is that a good deal? How many of you can say you know the vague price per ounce of palladium off the top? Of course if the guy gives you 5 minutes to look up the price of palladium you’d see an ounce currently goes for ~$1,000 and you’re getting the deal of a lifetime. What if the guy gives you 10 seconds? I’d guess many would pass as they have no idea what palladium is, what it’s for, and how much it costs.
Now imagine the same street palladium vendor, but this time he doesn’t just approach you, he approaches you and your group of 10 buddies, one of which is a commodities trader. You get offered the palladium for $100, your buddy rips your wallet out and hands the guy the $100, and you continue enjoying your night up $900.
The bias to stay away doesn’t make sense, as markets are likely MOST efficient when the information environment is very stable and easy to digest. If everybody on the street was a palladium expert, you likely would have never been offered that deal. Time crunch can make even trivial problems really hard, which opens up the opportunity to outperform.
Specific information can situationally be extremely valuable. My wager is that we’ll increasingly move towards two options.
Computers and AI
Consensus markets
If you think about it, this already how things are done. Palladium isn’t actually traded on the street, it’s traded via future’s contracts on exchanges with likely millions of hours of labor per month aggregated together to create a liquid market expressing consensus on price.
So we have these extremely liquid markets with super high participation, with I’d bet the majority of volume exchanging hands between parties who will never in their life even see physical palladium! Yet when it comes to “Is this $300b market cap health insurer an imminent $0 off criminal fraud?” the best we can do is a couple late sell-side notes and a mention on CNBC. The information market there I’d wager is extremely inefficient.
If we could rank everybody who knows anything about UNH or any stock really, I’d be quite surprised if the person who knew the most was in a position to deploy billions of dollars into the stock. I’d also be quite surprised if the most knowledgeable person had a better mental model than if we could combine the model of everybody else.
Say you had access to a magic computer terminal with the full contents of everyone’s brain uploaded to it. Any question you asked, it would provide the best approximation to reality. The WSJ article comes out and you could simply ask “Is this true?” and immediately get the answer, letting you front run the company announcement. Even then, unless you have a few billion laying around, it’s entirely possible you’re missing upside as others start to get the idea.
Of course there’s a lot of ground to cover between the current status quo and an index of all human knowledge, but I’d guess we trend in that direction. Consensus markets and AI seem like they have a lot they can contribute there, especially a combination thereof. Hence I’m not surprised Kalshi is partnering with xAI. Our own work with Clarity has inclined us towards weaving in Gemini models with our questions, and eventually loading context into specialized Gemini models for equity research. Context is everything with AI currently. I’d say we’re much closer to having the magical computer mentioned above compute wise, but it’s missing a LOT of information.
So let’s explore some information.
New Markets:
Not much to discuss this week on the new side. What we’ll likely try to end up doing is having a better way to explore this coverage consistently via our site. Some kind of dashboard type deal aggregating all the markets interesting for finance. In the meantime I’ll shotgun out some updates below.
Updates On Prior Markets:
US Q2 GDP:
Some slight declines in Q2 GDP expects.
CPI:
Inflation for April was marginally on the lighter end at 2.3%, bringing down forward expects slightly.
AI Ranks:
Gemini continues to strengthen their likelihood to end May as the best AI model. Their flash model now ranks #2 on benchmarks, above all OpenAI models and only below their pro model. I’d guess the price per token there is less than 1/10th the cost of OpenAI models of similar performance. I’m definitely a huge Gemini fan recently.
For more info on this, you can explore the ranking site here
Taiwan Invasion Risk:
Markets have shifted away from their idea of near term China x Taiwan risk. Odds for later in the year are lower liquidity and haven’t moved much, perhaps indicative of the relative uncertainty surrounding the Trump admin.
Tariffs:
Nobody cares about tariffs anymore, and expectations of trade deals outside of China have dropped across the board
June and July (and September?) Fed Decision:
The people still expect rate cuts, but increasingly less so day after day. It’s unfortunate there aren’t good markets on the deficit given sensitivities to rates.
Recession in 2025:
Odds still down a lot from peak tariff, but still much higher than peak Trump euphoria. I’d guess we are back to a good amount of euphoria, just now there’s a mix of higher Trump volatility included.
Conclusion:
Our future vision for Clarity Markets is to have a variety of single stock and market wide KPI’s that answer key questions for whatever you may want to know. This blog (and our website data subscriptions) can then be used to disseminate high signal consensus on what matters most.
Hopefully this article covered some interesting topics in a way that was helpful. If so, feel free to subscribe for more.