The Problem With Prediction Markets
There’s a classic joke that goes something like this: an economist is walking on the street and notices a $20 bill on the ground and nobody around who it belongs to. He examines the bill, and then walks away. The friend with him asks why he didn’t take the money for himself. The economist replied “if it was really free money, then someone else would have taken it already.”
Today we’re going to be talking about why prediction markets aren’t necessarily the solution to a lot of the problems we encountered last week in trying to figure out the issues with election polling and polling aggregation. The same as last week, this is meant to be a plain English, non-partisan explainer that goes over some of the issues with predictions markets. I think that this is a lot closer to the problems with markets (like the stock market) in general. Certainly some statistical lessons can be gleaned, but today I’ll flex some knowledge accumulated more from the finance parts of my day job.
The Basics of Prediction Markets
In many ways, finance is the art of transformation. When a company sells its stock in the private markets (typically Venture Capital or Private Equity) they are transforming a share in all future profits of the business into cash today. Similarly if I am a farmer and want to ensure that I have enough money to make it through to the next year, I might transform the future crops I will be harvesting from an uncertain future sum of cash (selling at whatever the price in the market will be) into a guaranteed price for a guaranteed amount of crops (depending on structure this is either a future or forward contract). In the latter case this also might be called a hedge - as in I am hedging against the risk that I receive a price lower than I expect and hope for.
In fact, markets and trading allow us to transform quite a few different things into dollar values. For instance, continuing on the crops example from above, let’s just assume that I’m locking in selling crops at the current market price today - call it $100 a bushel. So if the contract is to sell you my crops tomorrow at $100 (the current market price) then my hedge likely isn’t that valuable - how much could the crops change in price in the next 24 hours? But if the contract is to sell you my crops 1 year from today at $100 that could be very valuable indeed. After all, who knows what price the crops will be at next year? There could be a crop shortage due to poor weather. Maybe there will be an abundance next year on the other hand. Depending on a whole variety of factors, that uncertainty that time breeds will increase the value of locking in that price for someone, and decrease it for someone else. But the bottom line is that we’re transforming time into a dollar amount to be haggled over.
Continuing beyond time, we can also transform the credit worthiness of a counter party (the person you are doing business with) into a dollar value. If Microsoft decided to raise money to build out data centers, you would likely lend to them rather cheaply. This will be reflected in the interest rate that is paid. Now if your multiple-times-bankrupt buddy from college approached you to raise money to build out data centers, I am guessing you would charge him a higher interest rate than Microsoft.
These are basic and stylized examples, but the bottom line is that by using the same medium to express different more abstract concepts - creditworthiness, time, risk - we can turn them into concrete dollar values. That is the medium we are using to express our various opinions - and the value you place on time is an opinion in its own way - in dollars.
So markets and transformation to dollars give us a way to have a shared language for expressing and predicting things that otherwise might be impossible to clearly communicate to each other. We might have a completely different model for voter turnout, who women-under-35 will vote for, how to weigh the Quinnipiac Polls, and so on, but when I tell you that I am 40% sure that Jim Exampleson will win the election, suddenly you can either agree or disagree.
Efficient Market Hypothesis
With our shared language - percentages or dollars for now - we’re now in a much better place as we are no longer talking past each other. I might have a different model and a different prediction than yours, but so long as we’re using that same language - dollars or percentage - then we are in a coherent conversation with each other. I agree with your price or odds or I disagree and then try to profit off of that.
That’s something, but betting markets promise us more. In fact, this screenshot I took as I’m writing this from Polymarket - one of the larger prediction markets - promises “Live and accurate forecasts”
Let’s make a bold assumption for a second that this company actually does care about more than profits (bold indeed) and aspires to provide more accurate predictions. Why should prediction markets provide more accurate forecasts?
Well there are actually quite a few theoretical justifications for markets providing more accurate predictions. In fact, you’ll find some economists who defend price gouging during natural disasters because they see it as a price signal. Flights are more expensive out of an area because a hurricane is coming should signal to the market to provide more flights out of that pending disaster area1.
But let’s skip over justifications besides the profit motive and instead go to the end result of all of those incentives that markets provide (according to some): Efficient Market Hypothesis.
Efficient Market Hypothesis essentially states that asset prices reflect all available information. At a basic level the thinking isn’t too far from the joke I opened this piece with: if you’re buying a stock today, you (per this theory, if it holds true) should not be able to make any money through the efforts of your research into it. All of that information has already been incorporated into the stock price. What that does not mean is that you are guaranteed to lose money; there are bound to be random gyrations of the stock throughout trading periods. Furthermore, it is possible that the stock you bought has an unexpectedly strong quarter, and goes up as a result. Note, I said unexpectedly, because again in stronger forms of Efficient Market Hypothesis (EMH henceforth) all available information should be incorporated into the stock price.
Now from an outsider’s perspective if you haven’t heard this theory before, it can sound rather silly, hence the joke. But I would advise against dismissing it out of hand. Through a combination of legal constraint and other constraints, there are certain kinds of trades that you can reasonably expect to be positive that the market has not competed away. However, for the rest of the market when we discuss highly liquid (meaning trading frequently) stocks in major public markets, some version of EMH has solid explanatory power. After all, some of the smartest people in the world try to beat the market and often fail. Furthermore, when greater understanding of what drives prices are discovered - often described as and called factors such as bigger companies being more valuable, or companies that are focused on growth instead of maximizing value at current size, etc. - those factors tend to quickly lose their (superior) value.
Above I alluded to there being multiple forms of EMH. These forms will be based on how much information is hypothesized to be “baked in” to the price. A weak form EMH will posit only historic stock prices are incorporated whereas a strong form will speculate that even private information is included in the price information. Who knows, perhaps in the future insider trading laws are relaxed so much that becomes a reality!
How Does That Work Though?
So markets will lead to predictions that are in line with public data, and possibly even private data too. But how?
Continuing with EMH, we have a stock Example Industries LLC that is priced at $100. Everyone that owns it agrees that it is worth about $100 and as such it trades at $100 moving up or down a random bit each day but always closing around that $100.
Now Scrooge McDuck realizes that he’s missing out on profit by keeping his money in gold coins in a vault in his house and decides he’ll reinvest it. He goes to the stock market and decides he needs to get market exposure ASAP, prices be darned. So he begins by saying he’ll buy Example Industries LLC for $105. He’ll tell his stock broker to go to the market and convey that information; incidentally this would be called being bid $105. As an owner of Example Industries LLC I know that there’s a potential $5 worth of profit there (105-100=5) so I agree to sell at $105; this would be called being offer $105.
Great, right? However, we’re in EMH world, and everyone else in the market (somehow) knows that we should be selling this stock for anything above $100 as that would be a premium to it’s true value (if the assumptions hold). So the next guy says, “well anything above $100 is a premium, I’d feel good selling to Scrooge at $104.” Then the next guy says “well $103 is still profitable” and so on until Scrooge is buying for more or less about $100. Again, if EMH holds then we’ll all sell as that’s a premium.
In fact, this is a bit how markets work generally. There are High Frequency Traders looking to see if a stock selling at $101 on Exchange A can be bought for $100 at Exchange B. When that happens I will buy from B, and sell to A and lock in $1 profit. Everyone will do this until that price difference is eliminated. That guaranteed profit is called an arbitrage and is what will keep prices relatively the same across and between exchanges.
Using this language, we can think of EMH as a theory that all information in the markets will be arbitraged away. In the Scrooge McDuck example we all would have competed to sell him the stock at a premium until that competition ultimately drove the price down to ~$100.
Please note again that this is a highly stylized version of EMH and markets as a whole. Furthermore like I said there are $20 bills laying on the ground so to speak because of various restrictions that limit arbitrage. In fact, as a trader those are the opportunities you should be looking for. But also by that same logic, you should make sure you can understand why that arbitrage or opportunity exists, because chances are if you don’t understand it then you might be the Scrooge McDuck squandering money.
Back to Political Betting Markets - Less Liquidity, More Problems
So we now know that transforming to dollars will give us a shared language. Furthermore, we have a hypothesis that posits that in markets, new information will be arbitraged away. Sounds like prediction markets, and markets in general are the way forwards then?
Unfortunately there are limits.
Earlier I brought up, “when we discuss highly liquid (meaning trading frequently) stocks in major public markets, some version of EMH has solid explanatory power.”
Regarding the high liquidity, one way to characterize liquidity would be how much an individual trade or trader can impact the market. If we’re trading a penny stock (a cheap, generally lightly traded, and small company) and I buy $1mm (million, more finance speak) of it, that’ll likely push the price meaningfully upwards. If I do the same $1mm buy of Apple stock then there will be minimal if any price impact to that purchase. Hence we could refer to liquidity as a spectrum with some stocks being more liquid than others. As it is a spectrum it is a bit difficult to define how liquid the various prediction markets are, but I can tell you for a fact there are whales in those betting markets.
When I refer to a “whale” that essentially means somebody that is throwing around a lot of money. In the case of these prediction markets the whale represents about $30mm (million) worth of bets on Trump. Again as someone who works in public markets, $30mm is generally a pretty small amount if you’re talking an institutional or larger fund in the stock market, but these prediction markets are less liquid, and hence are impacting the prediction odds. I’ve seen reported that the impact of those purchases in particular have moved the odds ~5-8% more than they otherwise would have.
Again, when defining liquidity there’s no strict cutoff for how much any purchase should move the market, but I would say that if indeed $30mm worth of bets moves odds 5-8% ceteris paribus (meaning: assuming no other changes) then that hardly qualifies as “highly liquid”. So there we have our first issue with prediction markets; insufficient liquidity means that they can end up less reflections of all information and more reflections of large traders’ opinions.
Does EMH Hold for Prediction Markets? Experiment in 2024?
I think this is a much more academic and difficult question than people would want to accept. There’s a sort of critique of EMH that goes “well there was a big price move and thus EMH is wrong.” I think given that there tend to be few glaring opportunities for profit in the market and that many of those opportunities for profit can be explained by legal and size constraints for investment means we should treat EMH with more respect than simply dismissing it out of hand because it fails to explain every market move.
That said, to go with the more snarky dismissal we do have this:
To give a better justification than showing a time that prediction markets were big time wrong, let’s think about an investing philosophy that’s in contrast to EMH, generally.
Value investors tend to be folks that subscribe to an investment philosophy akin to what Warren Buffett preaches: by doing research into companies and figuring out which ones are good, you can be rewarded for holding those high quality companies over a long period of time. Sometimes those companies might suffer in the markets, but over time you’ll be rewarded for holding high quality companies.
If we extrapolate this to prediction markets and you can apply a value investing philosophy and be rewarded for it, then it would act as a repudiation to the idea that EMH holds in prediction markets and that prediction markets ought to provide a more accurate vision of events.
The thing is, this election looks to be a genuine 50/50 or so. Maybe it’s 55/45, maybe it’s even 60/40, but at least from my perspective, it looks like Kamala is favored in the popular vote, and Trump is doing alright in the swing states he would need. If we apply the logic from last week’s piece on polling, I’d put it more like: “based on some polling where they seek to correct for misses from the prior election and seek to cover ~95% of scenarios, then when we include the margin of error in polling at the current stage, this election looks like it should be somewhere between 40-60 favoring either candidate to win.”
Unfortunately, that doesn’t provide a neat falsifiable counter scenario for us. For instance, if the election polls had it Kamala 70/30, and Polymarket still had it 60/40 towards Trump, that would be perfect for us. In this case you could “value invest” based on the underlying polls, hold until the election when Kamala would be much more likely to win, and then cash out. Were that the case, it would be a potentially decisive blow that these markets aren’t efficient and more accurate.
Another Problem: Infrequent Marks to Market; Also, Prediction Markets’ Value?
A publicly traded company does just that: it publicly trades everyday that the markets are open. If I own a share of Apple stock, 5 days a week I will know exactly what that stock (and the larger company) is valued at.
Private Equity (PE) and Venture Capital (VC) are investment classes that focus on private companies. They will invest money in a company, and periodically if there is a large piece of news, or a material change in the value of the company they will update the value. For instance, we’ve heard frequently about how much in value Twitter has deteriorated since Elon Musk bought it and took it private.
Something interesting that has emerged in financial literature is that pensions, endowments, and various other long last institutional investors invest in and like PE and VC because they so infrequently update the value of the investments. We don’t need to get into the nitty gritty of it beyond the fact that it’s very convenient to not worry about updating values (for better or for worse) more or less daily.
Rather, I bring PE and VC up because these betting markets are more or less the same. There will only be one mark to market for these bets/predictions, and that’s when the bet settles and you either won or lost. Just as we can only know how good our election polling was after the fact, we’re in the same situation for our prediction markets.
Now that doesn’t mean they are worthless. In fact, they do have some value which is quite different than what the owners and boosters claim: these markets are liquid enough that they serve as useful barometers of the change in election odds based on recent events in real time.
To give a good example: after the debate between Trump and Biden (2024), Biden’s odds went down precipitously. That was a real time vindication of what we were seeing in terms of what looked like a poor performance. Before anyone answered a polling question about it, we could go to these various sites in real time and see that it augured poorly for Biden. That was borne out by history too (him dropping out).
But note that what I’m claiming here is starkly different from what these sites advertise. Real time prediction markets can be useful for gauging how the larger populace might be reacting to the latest news. That is a far cry from those markets being the most accurate, or whatever is being claimed.
EMH Failed to Account For:
People being stupid.
I’m being a bit reductive here for a punchline above, but the general thrust of this piece is related to prediction markets on politics in particular. We have quite a few different phrases to convey “someone is letting their emotions dictate their gambling” such as being on tilt, throwing bad money after good, etc.
Letting your emotions dictate how you allocate capital as opposed to a sound and repeatable process is - and this is my opinion here - stupid. And I promise as someone who has followed these kinds of markets (and other irrational markets) for a while that you could classify many of these traders as dictated by emotion rather than rational analysis. Politics does that to us!
At the end of the day we have countless examples beyond prediction markets of folks letting tribalism, memes, or excessive opportunism separate themselves from their money. Bed Bath and Beyond was raising capital from retail investors more or less till the day they went bankrupt. A few years ago Hertz announced it was essentially bankrupt but was still going to sell shares in the company despite that. People were lining up to buy stock in a self-admitted essentially worthless company until a court order halted that.
I said earlier that EMH actually does have decent explanatory power. You probably aren’t going to discover something looking through Apple’s various financial statements that will allow you to make superior returns on investing in it. But when we get to penny stocks or prediction markets? Suddenly markets aren’t so efficient, and the Hypothesis part of Efficient Markets Hypothesis starts to become the key word. We as a people let that stupidity and emotion affect our decisions, and few places will you see that more than in these political prediction markets.
Conclusion
We set out to look at prediction markets and examine a few questions. What is the theory behind why prediction markets should give us more accuracy? I used mostly EMH to explain it, but the profit motive (i.e. there’s a price dislocation and you look to make money off of it) is of a similar type of thinking and provides a similar justification.
We discussed how there are whales in these markets, and the larges moves they’ve made show that the markets aren’t highly liquid because those moves impact price/odds so much. Furthermore, we talked about how with political prediction markets in particular when folks are investing based on emotions that makes it more likely that the odds we’re seeing reflect sober analysis.
But not everything is bad here; the challenges above might affect the predictions as a whole, but there is still some utility in seeing the change in those flawed odds to get a sentiment check on how news affects candidates odds.
At the end of the day, we might not get a clean example to show why prediction markets aren’t necessarily more accurate, but I do think we saw why as they are currently there are structural problems that mean they aren’t as accurate as the websites portray them as.
Unfortunately there are only so many flights that can physically happen - so after a certain point due to constraints like the number of feasible flights with the infrastructure to safety concerns this price gouging fails to induce more supply and serves purely to capture consumer surplus, and from the richest people too, enriching airline executives but not enhancing welfare for people