I think your article is full of non-sequiturs. For example, top profiting accounts being bots on sports and crypto markets seems irrelevant to me, you should just be filtering those out because those are not what prediction markets are really about.
I make the falsifiable claim that if you are actually very good at forecasting, you will make a lot of money on prediction markets. Millions of dollars. Why pundits don't do this is a great question, you should let us know since you are one yourself! My answer is that it's because forecasting is actually very hard and the skill required to be a popular pundit is not highly correlated with good forecasting.
I think the issue is less that you can't make a healthy profit if you're good at forecasting (which, surely, one can) and more that prediction markets aren't as revolutionary a tool as one might hope for generating information about the future. Idiot hobbyists making ridiculous bets simultaneously make it easier for serious forecasters to turn a profit and make the prediction market less useful for a disinterested observer who "merely" wishes to use it as a tool to predict the future.
Interesting article, but it contains a couple of factual errors.
First, Kalshi (and Polymarket sometimes) pay interest or yield on funds tied up in eligible positions. That does not get rid of the opportunity cost of holding a long-term contract, but it makes the problem considerably less severe.
Second, bots do not dominate overall P&L. The methodology behind the report cited here is highly questionable: frequent trading is not the same thing as bot trading, and profitability among high-frequency accounts does not prove that bots capture most of the profits. Look through the actual P&L leaderboards; many of the largest winners are discretionary traders.
Bots clearly have an advantage in highly mathematical markets, particularly short-term crypto-price contracts. But extrapolating from those markets to claim that prediction-market profits are generally dominated by bots is not supported by the evidence.
Yes, but that is a much narrower claim. Bots dominating high-frequency trading does not mean bots dominate overall P&L. I don't think humans ever had a chance in the high-frequency crypto markets.
Why does it matter if bots dominate high frequency trading? I'm sure you would agree that the sports and crypto price markets are irrelevant to the societal value of prediction markets, those are where the bots are.
I know real people who don't even know how to code but are up millions of dollars forecasting things. It is possible and it is real. But they are not pundits and they usually have under 1,000 followers on Twitter.
Paying interest helps, but a real game changer will be when stock brokerages allow you to trade prediction markets on margin (without tying up any cash), like real futures. I'm shocked this hasn't happened already.
You can sell options on margin, where you can easily lose 10X premium overnight. This seems easier; the brokerage just needs to ensure the customer has enough margin to cover the contract going to zero, and they'll debit/credit your account after it resolves.
But then you would need margin equal to essentially the full potential loss on the position. A brokerage isn’t going to extend much credit to a customer whose entire portfolio consists of out-of-the-money options.
You could use a stock portfolio as collateral to fund prediction-market trading. But in that case, it’s the stocks providing the margin capacity, not the prediction-market contracts themselves.
There are a few startups I've heard of trying to address this problem but it's essentially unsolvable imo.
That's right, it would use stocks as collateral. Most people (excluding degenerate gamblers) have ample margin in their brokerage account to fund their predictions.
Alternatively: If you're betting more on prediction markets than the margin extended from your broker, you're probably a degenerate gambler.
AI forecasting firms will soon dominate prediction markets, much as quant shops already dominate trading in large financial markets.
The low cost of running LLMs compared to the time and attention of elite human capital dramatically reduces the amount of dumb money and/or market subsidies necessary for a market to arrive at the correct price.
This will make widespread use of prediction markets for important questions viable even if few smart people actually trade in them.
Without making a prediction on when or if it will happen, a world in which LLMs are good at forecasting will be a very different one to the one we have now.
I guess I am missing something. The prediction market is zero sum with an intended positive externality of creating and consolidating useful knowledge on future uncertain events. I agree that we need to eliminate the noise of stupid predictions on irrelevant events (will he make the next free throw?), but that doesn’t negate the benefits of generating important knowledge. The fact that bots or insiders participate is also meaningless.
The benefit of prediction markets isn’t in the game itself, it is the output.
On Metaculus forecasters being amateurs. As I know the Metaculus founder and second CEO fairly well, I have some insight. Their best forecasters on staff and in network often do work for hedge funds. Some hedge funds contract Metaculus directly to do forecasting work for them. And some of their individual forecasters do consulting work on the side with hedge funds. And de facto Metaculus acts as a farm system for those hedge funds.
And some of those that land a hedge fund job even still do Metaculus forecasting... because they like spending time with their friends. As for most of us, there is great pleasure in doing activities with other people who approach, think, and talk about the world in the way that they do.
I've been amused over the years that there are a small number of very "egghead" commentators (e.g. Scott Alexander) who are obsessed with prediction markets. Recently, I've been amused when prediction markets have become a real business -- exactly how any sensible observer of human behavior would predict, as casinos.
Some of this probably stems from asking Big Questions like "Do you think Democrats or Republicans are better for economic growth?" rather than real questions, like "Do you think Democrats or Republicans will improve the place of People Like Us on the class-status totem pole?"
BTW, that latter reminds me that Big Thinkers seem to overlook that humans often aren't happily cooperating to achieve common goals, where it is world socialism or infinite material wealth, but are often engaged in the inherently grubby business of competing with each other for status, a zero-sum game. It's as if they'd never read "The Selfish Gene".
Prediction markets need to focus on qs which are decision-relevant for policy in order to demonstrate to policymakers that they can have value. To do this, subsidize the accuracy of the prediction market on particular questions of relevance, and keep bets in escrow so they will grow with the stock market.
Interesting take. But one note, Polymarket actually pays interest on long terms bets. So they do try to reduce the problem of the cost of opportunity (not sure how effective it is so far, but it's a start).
I think your article is full of non-sequiturs. For example, top profiting accounts being bots on sports and crypto markets seems irrelevant to me, you should just be filtering those out because those are not what prediction markets are really about.
I make the falsifiable claim that if you are actually very good at forecasting, you will make a lot of money on prediction markets. Millions of dollars. Why pundits don't do this is a great question, you should let us know since you are one yourself! My answer is that it's because forecasting is actually very hard and the skill required to be a popular pundit is not highly correlated with good forecasting.
I think the issue is less that you can't make a healthy profit if you're good at forecasting (which, surely, one can) and more that prediction markets aren't as revolutionary a tool as one might hope for generating information about the future. Idiot hobbyists making ridiculous bets simultaneously make it easier for serious forecasters to turn a profit and make the prediction market less useful for a disinterested observer who "merely" wishes to use it as a tool to predict the future.
Interesting article, but it contains a couple of factual errors.
First, Kalshi (and Polymarket sometimes) pay interest or yield on funds tied up in eligible positions. That does not get rid of the opportunity cost of holding a long-term contract, but it makes the problem considerably less severe.
Second, bots do not dominate overall P&L. The methodology behind the report cited here is highly questionable: frequent trading is not the same thing as bot trading, and profitability among high-frequency accounts does not prove that bots capture most of the profits. Look through the actual P&L leaderboards; many of the largest winners are discretionary traders.
Bots clearly have an advantage in highly mathematical markets, particularly short-term crypto-price contracts. But extrapolating from those markets to claim that prediction-market profits are generally dominated by bots is not supported by the evidence.
It appears that bots do in fact dominate high frequency trading?
https://chatgpt.com/share/6a6df40b-bb18-83e8-9db2-7ad8186d44fc
Yes, but that is a much narrower claim. Bots dominating high-frequency trading does not mean bots dominate overall P&L. I don't think humans ever had a chance in the high-frequency crypto markets.
I don’t see why it even matters! What a stupid metric to be looking at in the first place (and I am not blaming you for this, just pointing it out).
Why does it matter if bots dominate high frequency trading? I'm sure you would agree that the sports and crypto price markets are irrelevant to the societal value of prediction markets, those are where the bots are.
I know real people who don't even know how to code but are up millions of dollars forecasting things. It is possible and it is real. But they are not pundits and they usually have under 1,000 followers on Twitter.
Paying interest helps, but a real game changer will be when stock brokerages allow you to trade prediction markets on margin (without tying up any cash), like real futures. I'm shocked this hasn't happened already.
It's not an easy thing to build. In traditional markets, assets rarely go to zero overnight.
You can sell options on margin, where you can easily lose 10X premium overnight. This seems easier; the brokerage just needs to ensure the customer has enough margin to cover the contract going to zero, and they'll debit/credit your account after it resolves.
But then you would need margin equal to essentially the full potential loss on the position. A brokerage isn’t going to extend much credit to a customer whose entire portfolio consists of out-of-the-money options.
You could use a stock portfolio as collateral to fund prediction-market trading. But in that case, it’s the stocks providing the margin capacity, not the prediction-market contracts themselves.
There are a few startups I've heard of trying to address this problem but it's essentially unsolvable imo.
That's right, it would use stocks as collateral. Most people (excluding degenerate gamblers) have ample margin in their brokerage account to fund their predictions.
Alternatively: If you're betting more on prediction markets than the margin extended from your broker, you're probably a degenerate gambler.
AI forecasting firms will soon dominate prediction markets, much as quant shops already dominate trading in large financial markets.
The low cost of running LLMs compared to the time and attention of elite human capital dramatically reduces the amount of dumb money and/or market subsidies necessary for a market to arrive at the correct price.
This will make widespread use of prediction markets for important questions viable even if few smart people actually trade in them.
Without making a prediction on when or if it will happen, a world in which LLMs are good at forecasting will be a very different one to the one we have now.
That world is already here - https://www.forecastbench.org/
Futa-rchic utopia? I mean damn it’s kind of a weird fetish but go off king. 👑
I guess I am missing something. The prediction market is zero sum with an intended positive externality of creating and consolidating useful knowledge on future uncertain events. I agree that we need to eliminate the noise of stupid predictions on irrelevant events (will he make the next free throw?), but that doesn’t negate the benefits of generating important knowledge. The fact that bots or insiders participate is also meaningless.
The benefit of prediction markets isn’t in the game itself, it is the output.
On Metaculus forecasters being amateurs. As I know the Metaculus founder and second CEO fairly well, I have some insight. Their best forecasters on staff and in network often do work for hedge funds. Some hedge funds contract Metaculus directly to do forecasting work for them. And some of their individual forecasters do consulting work on the side with hedge funds. And de facto Metaculus acts as a farm system for those hedge funds.
And some of those that land a hedge fund job even still do Metaculus forecasting... because they like spending time with their friends. As for most of us, there is great pleasure in doing activities with other people who approach, think, and talk about the world in the way that they do.
I've been amused over the years that there are a small number of very "egghead" commentators (e.g. Scott Alexander) who are obsessed with prediction markets. Recently, I've been amused when prediction markets have become a real business -- exactly how any sensible observer of human behavior would predict, as casinos.
Some of this probably stems from asking Big Questions like "Do you think Democrats or Republicans are better for economic growth?" rather than real questions, like "Do you think Democrats or Republicans will improve the place of People Like Us on the class-status totem pole?"
BTW, that latter reminds me that Big Thinkers seem to overlook that humans often aren't happily cooperating to achieve common goals, where it is world socialism or infinite material wealth, but are often engaged in the inherently grubby business of competing with each other for status, a zero-sum game. It's as if they'd never read "The Selfish Gene".
Prediction markets need to focus on qs which are decision-relevant for policy in order to demonstrate to policymakers that they can have value. To do this, subsidize the accuracy of the prediction market on particular questions of relevance, and keep bets in escrow so they will grow with the stock market.
Interesting take. But one note, Polymarket actually pays interest on long terms bets. So they do try to reduce the problem of the cost of opportunity (not sure how effective it is so far, but it's a start).
Doesn't Good Judgment project solve for this, it's just about bragging rights with no money down