Why Is the Demand for Prediction Markets So Low

I love the idea of prediction markets. I’ve loved it since the first time I learned about it. Just ask a question and rely on the magic of market optimization to get the best available approximation to the answer.

I’ve had some experience participating in prediction markets. In 2020 I earned around $10k on Polymarket betting on the US election outcome. Last year I created an account on Manifold and since then created around 90 questions about space, science and random stuff.

Yet, with all that I don’t think that prediction markets in their current form will take over the world or for that matter be widely useful except in some limited use cases.

I’ll structure my reasoning in two parts: from the perspective of a trader and of the question author. The reasons are somewhat reflective of each other, but I think it’s still beneficial to look at them from two points of view.

Trader’s view

Suppose you have a spare $1000 that you can invest in either stock market or prediction market. What are your options?

If you don’t trust your own judgement, then you can’t do better than investing in S&P 500, which will on average produce 7-8% returns (adjusted to inflation).

If you are an expert in some field and/or have some insider information, you could buy stocks in a company that you think will grow faster than the market as a whole. This could bring you significantly higher returns. Even if you are wrong about being an expert or about your insider information, chances are you will still win, since stock market is a positive-sum game.

Now let’s look at prediction markets. Under which conditions do you stand to gain more on prediction markets than on the stock market? Unlike the stock market, prediction markets are mostly¹ a zero-sum game. For a year-long prediction market the only way for you to get higher returns compared to investing in S&P 500 is to have a probability estimation for the outcome which is better than the market by 8%. For comparison, hedge funds generally have less of an edge in profitability over S&P 500.

How can you gain this 8% advantage? I see several options:

1. Catch the market right after it has been created or after some event has drastically changed the probability.

2. Have some insider knowledge.

3. Do the work: build a better model, or do better research than other traders.

All of this requires either work or knowledge and has to be adequately compensated, which would only work in markets with enough liquidity.

All of the liquidity in the market is added either by the market creator, or by other traders. We’ll talk about the creator’s view in the next section. When it comes to other traders, if we assume that they act rationally, all of their bets should also be made with the expectation to win on average. However the total wins in a single market are limited to the initial liquidity provided by the creator, meaning the total expected wins should be likewise limited. (Strictly speaking the total expected wins for a single market can be higher than the available subsidy, but when aggregated over all markets, the total expected wins shouldn’t exceed the total subsidy if all of the traders are acting rationally.)

The situation with redistribution of bets therefore is more similar to professional poker, than to stock market: the gains of the better players/traders are financed not by the stocks growing on average, but by the bets of the less skilled participants.

Market creator’s view

Suppose you have a spare $1000 and you want to spend it on learning the answer to a question X. What are your options?

You could create a question on a prediction market and provide $1000 worth of liquidity, or you could hire an expert and pay them $1000 to research the answer for you.

For simplicity let’s assume that question-answering efforts can be quantified as “hours spent researching”. Hiring an expert gets you exactly $1000 worth of research-hours. In comparison to this, how many research-hours does a $1000 prediction market buys you?

First, suppose there is a relatively clear answer to your question and the market quickly settles on the objectively correct probability. How much do the traders in the market stand to gain from this prediction? Suppose you provide the initial liquidity by investing $1000 in YES, $1000 in NO. Out of this initial liquidity, other traders can earn up to $1000. So, superficially you can get out of the prediction market almost the same $1000 worth of research-hours.

However there are several additional costs:

• You need to tie up an additional $1000 until the market resolves.

• Other traders also need to tie up some of their money in the market, which reduces their effective profit, and hence the number of research-hours that you buy.

• Multiple traders will need to do the same bits of research. At the very least all of them will need to spend time reading the market description. This means that some of the research-hours that you bought will be spent redundantly.

Based on this, the number of non-redundant “research-hours” that you buy via a prediction market should generally be significantly lower than if you buy them directly from an expert.

So, what advantages can prediction market bring to account for this?

Some of the traders that participate in your market might have very specific expertise or insider knowledge that will allow them to give a precise answer by spending much smaller amount of time than an expert that you would otherwise hire.

It is possible that people betting against each other will thereby inflate the amount of available liquidity. For this to happen, the question has to be divisive, and should provoke at least some of the traders to act irrationally, i.e. to gamble. If all the actors in the markets act rationally, their expected gains should total to the amount of provided liquidity.

So, where does it leave us? To win over a hired expert, a prediction market should pose a question that either requires very particular knowledge, or should be interesting and divisive, inducing many traders to bet based on their biases.

What does it mean for Manifold?

Manifold, as well as Metaculus have a significant advantage over theoretical abstract prediction markets: they are gamified and have a strong community of people that care about predictions. Just a couple of weeks ago I spent much more money on travelling to take part in Manifest than I could reasonably expect to ever earn on Manifold.

This reliance on the community making predictions for the sake of having fun makes it possible to ask questions like whether Grabby Aliens are real, or when Betelgeuse will go supernova, even though those questions could never be strictly profitable in a traditional prediction market.

Still, if Manifold wants to double down on the purely economic side of things, one way to do it would be to artificially grow all of the mana invested in the markets. This would make it profitable to a) bet on the longer-term markets, b) bet while having only a small advantage over the market.

Regardless of whether Manifold implements this suggestion, I personally will continue participating, since this is one of very few platforms giving quantifiable answers to interesting questions.

1. Depending on how subsidized is the market by the creator.